WPS4377 Policy ReseaRch WoRking PaPeR 4377 Early Identification of At-Risk Youth in Latin America: An Application of Cluster Analysis Emilie Bagby Wendy Cunningham The World Bank Latin America and the Caribbean Region Human Development Department October 2007 Policy ReseaRch WoRking PaPeR 4377 Abstract A new literature on the nature of and policies for youth in Mexico are suffering the consequences of a range of in Latin America is emerging, but there is still very negative behaviors. Another 8 to 20 percent demonstrate little known about who are the most vulnerable young factors in their lives that pre-dispose them to becoming people. This paper aims to characterize the heterogeneity at-risk youth ­ they are the candidates for prevention in the youth population and identify ex ante the youth programs. The analysis finds two observable variables that are at-risk and should be targeted with prevention that can be used to identify which children have a programs. Using non-parametric methodologies and higher probability of becoming troubled youth: poverty specialized youth surveys from Mexico and Chile, the and residing in rural areas. The analysis also finds that authors quantify and characterize the different sub- risky behaviors increase with age and differ by gender, groups of youth, according to the amount of risk in their thereby highlighting the need for program and policy lives, and find that approximately 20 percent of 18 to 24 differentiation along these two demographic dimensions. year old Chileans and 40 percent of the same age cohort This paper--a product of the Latin American and Caribbean Region , Human Development Department--is part of a larger effort in the department to understand the challenges facing at-risk youth in the region and to design effective policy to support them. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at Wcunningham@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team EARLY IDENTIFICATION OF AT-RISK YOUTH IN LATIN AMERICA: AN APPLICATION OF CLUSTER ANALYSIS Emilie Bagby and Wendy Cunningham1 1University of Illinois Urbana-Champaign and World Bank, respectively. I. Introduction The specific needs of young people are an increasingly debated subject in the development community. A large literature has emerged to inform the debate in Latin America, principally by mapping youth behaviors, positing policy and program interventions to address the behaviors, and, to a lesser extent, to identifying the (economic) factors driving young people's decisions (Lloyd 2005, World Bank 2006). A primary shortcoming of the research is that it treats youth as a homogenous group and reports average behaviors, thereby not capturing the complexities of the youth population across its many dimensions. Even if we do understand the heterogeneity in the youth population, appropriate policy requires an understanding of why young people make the decisions they do. The evidence for Latin American youth is limited. The role of economic incentives and budget constraints affecting decision-making by young people in developing countries has been examined (World Bank 2006 and the sources within), building on Gruber's (2001) work in the United States. A second line of work considers the broader context in which young people form their preferences and make their decisions (World Bank 2003, 2007), building on the public health and psychology research in the United States. While this literature gives good insights as to how to modify behaviors, it is silent on whom prevention programs should target. This paper aims to construct a more complete picture of the heterogeneous youth population in LAC by quantifying and describing the at-risk youth population and by identifying easily observable variables can be used to identify, ex ante, those youth who have a higher propensity of engaging in risky behaviors. We use non-parametric methodologies and special cross sectional youth surveys from Mexico and Chile that permit us to identify the factors that today's risk-taking youth had in their childhoods, which are then used to identify who should be the target of prevention policies and programs.2 The paper finds that over 20 percent of 18 to 24 year olds in Chile and 40 percent in Mexico have influences in their early and current lives that predispose them to negative behaviors, they have engaged in these behaviors, and they are suffering the consequences. Conversely, 40 percent of Chileans age 18 to 24 and 16 percent of Mexicans of the same age do not display any of these risks, behaviors, or consequences. While many of the factors that identify if a person will engage in risky behavior are unobservable, we find two observable variables that are highly correlated with these underlying factors and can be used 2Causality can only be determined using panel data that includes information on behaviors and household factors, which do not exist in LAC, to the best of our knowledge. 2 to identify which children have a higher probability of becoming troubled youth: poverty and residing in rural areas. While self-identifying as indigenous is correlated with risky behaviors, the relationship is not statistically significant. And, we find that risky behaviors increase with age and differ by gender, thereby highlighting the need for program and policy differentiation along these two dimensions. There are five sections following this introduction. Section II presents the conceptual framework and provides a brief review of the literature. Section III describes the methodology and Section IV discusses the data. The results are presented in section V and section VI concludes. II. Review of the Literature and Conceptual Framework The issue of youth development is relatively new in the field of economics. Perhaps the most extensive study in the economic literature is Gruber (2001), which investigates the determinants and implications of nine different behaviors among US youth ­ smoking, risky driving, sexual activity, suicide, marijuana use, crime, alcohol use, school dropout, and mis- nutrition ­ using both cross sectional and time series data. The rational addiction model developed by Becker and Murphy (1988) is expanded to allow for youth-specific characteristics identified in the psychology and human development literature, namely myopia, time inconsistent preferences, and projection bias.3 The book concludes that youth respond to incentives such as age-specific legal penalties, prices, and income and that the marginal cost to additional risk taking is small once participation in risk behavior has begun. Recent studies have tested these conclusions in the context of developing countries, finding similarities and differences with the US4. A shortcoming of the economic literature is the absence of a discussion about preference formation and that early experiences vary across the youth population, leading to heterogeneity in decision-making during the youth years.5 In contrast, an extensive literature in the public health and psychology fields start from the assumption that preference formation and constraints before the youth years, as well as during them, explain a significant portion of the variation in youth behaviors. The ecological risk framework posits that youth are a product of individual (personal), micro-, and macro-environmental factors (Bronfenbrenner 1979). The individual factors are those skills, behaviors, and ideas that are 3"Projection bias" is understood as today's preferences may not be representative of future adult preferences. 4See Lloyd ed. (2005), World Bank (2006), and Attanasio et al. (2005) for example. 5Gruber (2001) discusses preferences, but it focuses on the time-inconsistency of preferences between the youth period and adulthood. It does not investigate why different youth have different preferences. 3 "hardwired", rather than formed, such as rage, optimism, or general health. The micro factors include preferences taught and formed by the family, peers, community, and local institutions and the constraints imposed by the same, including household poverty. The macro factors include more general influences and constraints, such as gender/race discrimination, armed conflict, poverty and economic inequality. These factors are commonly classified into two groups: the set of personal, macro, and micro factors that increase the risk of negative behaviors (risk factors) and the set that prevent youth from engaging in negative behaviors, commonly called protective factors. Each person has a set of risk and protective factors that influence preference formation, constraints, and thus behaviors. Most of the empirical testing of the ecological risk framework depends on correlations between current risky behaviors and personal, micro, and macro factors of the youth population. The literature in the US primarily focuses on identifying factors related to single risky behaviors (e.g. substance use).6 Recent US literature has begun to take advantage of longitudinal data available in the US7 to demonstrate that many of those factors correlated with risk taking behaviors are actually causal factors. The few studies that have tested the model using data from Latin American and Caribbean find similar results to those in the US. Blum (2004) uses data collected in 11 Caribbean countries and finds that a positive relationship with a caring adult, whether in the family or in school, is a key factor that is positively correlated with less risky sexual behavior (sexual debut and condom use), contraception, pregnancy and childbearing. Lloyd (2005) identifies the positive effects of schooling and health on transitions to risk-free adulthood 6There is an extensive literature in the US which is too large to cover here. The US National Library of Medicine and National Institute of Health maintains a web page (http://www.ncbi.nlm.nih.gov/sites/entrez), that lists hundreds of published articles that have used this methodology. A few of the papers that motivated this paper include: Resnick et al. (2004) look at risk and protective factors related to youth violence; Blum et al. (2002) describe the ecological risk framework and provide empirical evidence for three risky behaviors (weapon related violence, ever had sexual intercourse and ever used cocaine); Scal et al. (2003) look at risk and protective factors related to smoking; Zweig et al (2002) identify methods of predicting risk profiles using risk and protective factors such as psychosocial development, school and family characteristics using OLS and multinomial logit regressions with cross sectional data; and Bernat and Resnick (2006) provides a comprehensive review of the resiliency framework and additional empirical support for this framework in promoting healthy youth development. 7These studies, which also appear on the NIH web site, use the National Longitudinal Survey of Adolescent Health (ADDHEALTH) through the University of North Carolina, Carolina Population Center, surveys youth in grades 7-12 with the first wave of interviews in 1994 (http://www.cpc.unc.edu/projects/addhealth/design). Follow-up waves were in 1996 and 2002, the latter enabling more detailed analysis. This survey is school based and asks about risk and protective factors as well as behaviors. The National Longitudinal Survey of Youth (NLSY) 1997 from the Bureau of Labor Statistics, surveyed males and females born in 1980-1984 (http://www.bls.gov/nls/nlsy97.htm) and focuses primarily on educational and employment outcomes. The previous NLSY was from 1979. 4 using DHS surveys from developing countries. World Bank (2007) finds that low self- esteem, spirituality, school connectedness, abuse in the home, abuse in the community, connectedness to institutions, poverty and gender are the factors most associated with risky behaviors and negative outcomes in Brazil.8 Youth in the English-speaking Caribbean who have lower risky behaviors9 are also those who are more connected to parent/family, attending religious services, feel little rage, have not been abused or witnessed parental violence, do not have family members who have attempted suicide, have mentally healthy parents, and have households free of illegal substances (World Bank 2003). A series of papers by Brook, et. al. (2001, 2002a, 2002b) find similar correlations between household factors, poverty, and community factors for explaining substance use and violence among Colombian youth, while Hutz and Silva (2003) find that young Brazilian men who have been incarcerated for violence are disproportionately the sons of poor, uneducated, and violent fathers. The ecological risk framework can be stated more formally. A person i has a set of behaviors, Bi , that are determined by a vector of risk factors,ri , and protective factors, pi , determined at the individual ( Ii ), micro (ci ), and macro ( Mi ) levels. Bi = f (r(Ii,ci,Mi ), p(Ii,ci,Mi )) (1) If an element in any of the vectors Ii ,ci , or Mi leads to a positive behavior in Bi , it will take a positive value in p(Ii,ci,Mi )and a 0 in r(Ii,ci,Mi ). Likewise, an element that leads to risky behavior in Bi will take a 0 value in p(Ii,ci,Mi )and a positive value in r(Ii,ci,Mi ). A weighted average of the risk (ri ) and protective factors ( pi ) specific to each person will predict the behavior elements in the vector Bi . Behaviors include elements such as unprotected sex, school truancy, or substance use. The outcomes of these behaviors are a function of the behaviors, the individual, micro, and macro environments, and luck ( ). The outcome, Oi , is given by Oi = f (Bi, Ii,ci,Mi, ) (2) 8Behaviors and outcomes studied include: grade repetition, early labor force entry, early sexual initiation, risk taking sexual practices, alcohol use tobacco use, illegal drug use, violence and suicide attempt. 9Behaviors and outcomes studied include: perception of general health, ever had sexual intercourse, ever attempted suicide, violent behavior, problems due to alcohol and drugs. 5 Outcomes can be good (school completion, youth participation) or negative (school dropout, exclusion). Risk and protective factors are included in the function since they can magnify or mitigate the outcomes of a behavior. We assume a distribution of that is constant across people, but instrumental in determining if behavior Bi becomes outcome Oi . Equations 1 and 2 can be used to link the concepts of risk/protective factors, behaviors, and outcomes to levels of risk and give insight to policy. A person with high values of the elements in r(Ii,ci,Mi ), low values of the elements in p(Ii,ci,Mi ), and a Bi and Oi that displays few negative behaviors and negative outcomes, is defined as being type I risk. More generally, a person classified as type I risk has many risk factors and few protective factors, indicating that they have a predisposition to engage in negative behaviors, but the person has not undertaken any risky behaviors. Prevention programs would be targeted to this group.10 A person with high values of the individual elements in r(Ii,ci,Mi ), low values of the individual elements in p(Ii,ci,Mi ), a Bi that displays many negative behaviors, and an Oi , with few negative outcomes would be classified as type II risk, where risk factors are present, protective factors are few, and the young person has engaged in risk-taking behaviors without having experienced any negative consequences. These youth are at-risk of suffering consequences and can thus benefit from prevention or second chance programs. Youth with positive values in the Oi vector are those who are suffering the consequences of their behaviors and are thus categorized as type III risk. These youth tend to have high risk factors, poor protective factors, and many negative behaviors, many of which might be identified prior to their suffering the consequences of their risky behaviors. They are candidates for second chance (remedial or rehabilitation) programs. III. Methodology To understand heterogeneity in the youth population, we use cluster analysis to identify different groups of youth based on the observable elements in the vectors Ii , ci , Mi , Bi , and Oi . Classifying these groups along the risk profiles will allow us to measure the share of the youth population that is at different levels of risk and to identify observable 10See Blum (1998) for a discussion of resiliency based intervention programs. 6 factors that can be used by policymakers to target programs to youth with type I, II, or III risk. A. Cluster Analysis Cluster analysis is a means to identify correlations across large data sets without imposing, a priori, a structure on the data. Observations are grouped based on minimizing a distance measure between each variable for each observation, i.e. the observations in a cluster share a set of common variables. By comparing the mean values of various variables across clusters, we can characterize each cluster. Ward's method (minimum-variance) of testing was selected since it provided the most distinct and interpretable clusters.11 Ward's method uses the error sum of squares criteria.12 The variance is minimized by calculating the sum of squared errors from the mean of the cluster for each of the m variables for each observation: W = ( xijk - xjk )2 (3) k j i i = 1,...., n observations, j = 1,...,m variables, and k = 1,...,l clusters Initially, each of the n observations forms its own cluster. The first merge is identified by calculating the sum of squares for each pair of cluster. The pairing with the smallest sum of squares is identified and those clusters are joined, leaving n-1 clusters. The second grouping calculates the sum of squared errors again and pairs the two clusters that have the smallest value, leaving n-2 clusters. The process is repeated until the optimal number of clusters has been reached. Three tools were used to determine the optimal number of clusters. First, stop commands following two possible rules (Calinsky-Harabasz and Duda-Hart) were used to find criterion for each cluster possibility.13 The Calinski and Harabasz method suggests the optimal number of clusters (g) that maximizes an index C(g) which uses the pooled within- cluster covariance matrix (W) and the between-cluster covariance matrix (V), where C(g) = [trace (V)/(g-1)]/[trace (W)/(n-g)] (4) The Duda and Hart method maximizes D(g), where 11There are many different ways to perform cluster analysis, and no particular method is considered the best. Ward's linkage cluster analysis is a commonly used agglomerative hierarchical method. 12An attractive feature of the Ward's method is that it performs well with groups that are of unequal size, which, as will be shown in the results, strongly characterizes the data. See Everitt et al. (2001) for a theoretical discussion of Cluster Analysis and Ward's criterion. See Cunningham and Maloney (2001) for an application. 13These two methods are implemented in STATA. They were identified as the two best methods available (out of 30) by Milligan and Cooper (1985) and are discussed in Everitt et al. (2001). 7 D(g) = Je(2)/Je(1) (5) Je(2) is the sum of the within cluster sum of squared distances between the objects and centroid if the cluster is split into two and Je(1) is the within cluster sum of squared distances. The local criteria calculated in equations 4 and 5, C(g) and D(g), are then combined with test statistics for each clustering option to suggest the optimal number of clusters. Larger values of both methods indicate that the clusters are more distinct from each other while lower values indicate that the clusters are not very different from each other and therefore artificially sub-divided. Second, dendrograms were used to select among the multiple "right" clusters that the other methods may give. Dendrograms graphically depict the hierarchical relationship between the clusters by showing the order in which clusters are merged as well as the distance between the clusters. At each level of the cluster formation process, a dendrogram can be generated to view the relationships between the clusters. The dendrogram changes as clusters are grouped and un-grouped, thus enabling the researcher to optimally choose the clustering level. Third, once the optimal number of clusters is suggested and the dendrograms generated, the clusters themselves are observed and the means of the variables are compared across clusters. The differences found between the different clusters are used to ultimately determine the optimal number of clusters. For instance, if 5 clusters were suggested, 6 clusters were investigated to see if there was an interpretable difference. If not, 5 clusters were used, however if so, 6 clusters were used. This process could then repeat. The cluster analysis is performed for 10 cohorts, identified by age, gender, and country. The sample for each country is divided ex-ante by age and sex since, when pooling the sample, these two variables dominated the clusters to such an extent that the risk and protective factors of interest played a very small role. Six cohorts from Mexico are analyzed: female ages 12 to 14, female ages 15 to 17, female ages 18 to 24, and males in each of the three age groups. Only the four older cohorts from Chile are analyzed since youth age 12 to 14 were not included in the sample. While the objective of a cluster analysis is to identify which variables move together, a decision was made to treat some variables endogenously and others exogenously. For example, a hypothesis is that poverty status is a good indicator for a youth being "at- risk". If we use this variable to create the clusters, it is possible that poverty is such a strong factor that it drives the clusters and renders the other variables meaningless. Thus, for these 8 type variables, we carried out the analysis treating them as both endogenous and exogenous and found little difference. We thus report only the results for treating them exogenously. The advantage of cluster analysis is that the only priors required are in the variable construction, such that they range between 0 and 1. For continuous variables, the value was normalized: x = y (6) ymax Binary variables were assigned a 0 or 1 and discrete variables were assigned a value between 0 and 1 based on the ordering of the responses. A variable takes a value of 1 the closer it is to the variable being described. For example, the variable "abuse" takes a value of 1 if there is abuse in the household and a 0 if there is not while the variable "connected" takes a value of 1 if the respondent reports that they reach out for help all of the time if they have problems, a value of 0 if they never reach out, and a value in between depending on how frequently they reach out. Variables that could not be ordered in a logical way were not included in the analysis. B. Identification of Correlation between Factors and Behaviors Using the generated clusters, we use parametric methods to test the relationship between behaviors/outcomes and factors. First, we graph the clusters using bubble graphs to show the relationships between risk and protective factors with positive behaviors14 then we regress the average level of positive behaviors from each cluster against the average level of risk factors and protective factors from each cluster, 15using OLS to identify if there is a significant relationship between the behavior and the risk or protective factors. We consider the relationships in both countries for each age cohort and for the sample as a whole. We also test the correlation between positive behaviors and three easily observable variables that seem correlated with good behavior: low poverty, ethnic majority, and urban residence. 14The size of the bubble represents the sample size for each cluster within the age/sex groups. 15For this exercise, we need to assign a "good" or "bad" subjective ranking to each I, c, M, B, and O. This is straightforward for most cases, but in three instances, the variable changed value from 0 to 1 and vice versa depending on whether the youth is younger than 18 years of age or not: parenthood, working, and marriage. The case can be made that for those under 18, it is not optimal to be a parent, to be working, or to be married. Indeed, this opinion is supported by the large body of research attempting to explain how and why youth arrive at these adult outcomes at an early age. At the age of 18, however, it is inappropriate to consider a person with these characteristics to be necessarily at-risk. Therefore, for those under 18 years of age, the indicators for being a parent, being married and working take a 0, whereas for those 18 to 24 the indicators take a value of 1. 9 IV. Data Description A. Data As youth departments and governments become more sophisticated in their efforts to understand youth, several LAC countries have developed specialized youth surveys. We use two surveys in this paper: the 2003 National Youth Survey (Encuesta Nacional de Juventud) from Chile and the 2000 National Youth Survey from Mexico. Youth are defined as being between the ages of 12 and 24 for this analysis, consistent with World Bank (2006). Chile first implemented their National Youth Survey in 1994 and has subsequently repeated it every three years. This paper uses the most recent data set that could be accessed, which was collected in 2003. A nationally representative sample of youth aged 15 to 29 is surveyed, reaching all regions (urban and rural) in the country, with a sampling error of 2.1% and a 95% confidence level. The sample size for youth aged 15 to 24 used in the analysis is 532116. The survey questions cover socioeconomic status, youth behaviors and opinions, and family backgrounds. It includes many questions about risk factors such as family cohesion, neighborhood violence, and social exclusion. Many protective factors are not covered, but there is information about trust in institutions, connectedness, and relationship with parents. The behaviors and outcomes are limited to employment, schooling, sexual health, victimization, discrimination and participation in activities. It does not include questions about whether the respondents take drugs or commit violent acts. There are very limited health and health access data. Mexico carried out its National Youth Survey in 2000, through the National Institute for Statistics, Geography, and Information (Instituto Nacional de Estadística, Geografía e Informática, INEGI).17 The survey sample is nationally representative and was performed in two stages: the entire household was surveyed and asked basic household characteristics, and, at a later date, youth aged 12 to 29 were asked a set of youth-specific questions. Only those aged 12 to 24 were included in this analysis, resulting in a sample of 37,979 respondents18. 16For this analysis, 195 observations out of 7,189 (2.7% of sample) for youth aged 15 to 29 were dropped due to missing data. We then restricted the sample to youth aged 15 to 24 dropping another 1674 observations. In many cases, missing responses could be coded based on responses to related questions so as to maintain a larger sample size. For instance, if someone does not respond as to whether they attend church, after they have already indicated that they do not believe in God, then we assume that they do not attend church. It would be useful to do analysis using censored data tools. 17The survey was repeated in 2005, but the data was not available at the time this paper was under preparation. 18Of the almost 60,000 youth respondents, about 10,000 youth aged 12-29 were not surveyed the second time, and were dropped from the sample used for the analysis. The reasons for not interviewing these youth were 10 These data are similar to those from Chile; however they are slightly stronger on protective factors and more limited on risk factors. The Mexican dataset does not have information about connectedness with other adults, abuse in home, victimization or violence in community, but it does have questions on attitudes towards drugs, the number of parents in the household and parental response to good/bad behavior. As with the Chile survey, there is information on individual earnings and three family poverty variables ­ parental earnings, parental education and durable goods in household ­ as well as regional/community level information. Although youth aged 12 to 29 were surveyed, those aged 12 to 14 were not asked questions about sexual activity or practices. To the extent possible, similar variables are used in our analysis for both countries, although the Chilean data are more robust with the protective factors and the Mexican data include more risk factors. Table 1 describes each variable and identifies which variables are used in which clustering exercise and which are treated exogenously. B. Descriptive Statistics Descriptive statistics for continuous and binary variables are provided for risky behaviors and factors in Table 2.19 Sample means for all of the normalized variables are presented in Annex A to provide a baseline of the average status of the population against which to compare the variable values in each distinct cluster. The Chileans are 1.7 years older in the sample than the Mexicans, but this is primarily due to the inclusion of 12 to 14 years olds in the Mexican survey. Thus, it is important to compare characteristics by age group, as presented in Table 2. Both surveys are slightly more weighted toward women. About 11 percent of the Chilean sample self-identifies as indigenous; ethnicity was not tracked in the dataset: they did not want to participate, were not at home at the time of the interview and would not return within the week, were on vacation, were working or at school in another city, were disabled, and other. Comparing the poverty variables (education level and monthly earnings of heads of households) and rural means of this dataset before and after dropping the data showed no significant difference at =.01 (However, at =.05, the difference between the education level of heads of households the original and remaining datasets is not zero). An additional 5% of the observations were dropped in creation of the variables. Since this was such a small amount of the sample, the resulting dataset was not statistically different from the original. Finally, the data were restricted to youth aged 12 to 24, thus further decreasing the sample size by 8903. 19Variables with categorical responses cannot be summarized, but the mean of the variables created from the categorical responses can be. Thus, the categories are not summarized in Table 2, but the mean of the variable created from the categorical responses is summarized in Annex A. 11 identified in the Mexican survey. In Chile, about 13% of the sample is considered rural20 while in Mexico about 24% is rural. In both countries, a significant percentage of youth in both countries are dropping out of school early and not working after age 18. In the survey, 29.9% of Mexican and 14.2% Chilean youth have dropped out of school prior to completing high school, but 41% of Chileans age 18 to 24 and 27.5% of Mexicans in the same age range are in school.21 While 20 to 30% of Mexican and Chilean youth are inactive after age 18, 12.8% of Mexican and 6.7% Chilean youth are inactive before age 18. Chileans have earlier and riskier sex than Mexicans. While only 13% of Mexicans age 15 to 17 report having had their first sexual experience, 27% of Chileans in this age group report the same. Chileans are 10 percentage points less likely to use condoms than are Mexicans, which reflects that 2.5 percent of Mexicans and nearly 4 percent of Chileans age 15-17 have had their first child. Marriage rates in Chile are less than one-third of those in Mexico for each age group.22 In terms of the risk and protective factors at the micro-level, Mexican and Chilean youth report a high level of personal connections with someone. Mexican youth report a wide range of topics that they communicate with their parents about, from school and work to politics and religion. While Chileans report good relationships with their parents, ten percent of Chileans also note important relationships with other adults. On the negative side, six percent of Chileans have suffered abuse in the home and 8.5 percent report substance abuse in their homes. Mexican youth tend to trust government institutions less than they trust local institutions. They feel that school quality is good on average, although half say they do not have access to health services. There is a connection to religious institutions as 21 percent of Chilean youth and 9 percent of Mexican youth attend church weekly, with 66 percent of Mexican youth attending church at least once in the past month. Of the questions on individual risk/protective factors, more than 85 percent of Chilean youth are optimistic about the future, similar to rates in Brazil (Instituto Cidadania 20The sampling was done in communities with at least 2000 inhabitants, so a rural indicator means that the respondent comes from a community with between 2000 and 5000 inhabitants. An urban respondent lives in a community of at least 5000 people. 21If there is no repetition, youth would finish high school at age 17 or 18 in both countries. 22In Chile and Mexico, the legal age of marriage age is 18, but Mexican youth can marry as early as age 16 with parental consent. 12 2004), and an even greater share, 98%, of Mexicans have a sense of well-being. However, half of Mexican youth and one-fifth of Chilean youth feel socially excluded. There are few questions on macro factors, but the data do show that over half of the Mexican youth are from families whose parents have less than a primary education, while over one-third of Chilean youth have parents with less than a primary education. In addition, over half of Chilean youth have felt discriminated against at some time. V. Results The clustering methodology identified four to seven sub-groups within each of the 10 age-sex-country cohorts, resulting in 53 (23 from Chile, 30 from Mexico) distinct groups of youth. To simplify the discussion, the commonalities across the clusters are presented in this section. The results are presented by risk category to allow us to characterize the heterogeneity in the youth population and to create a picture of youth in each category, based on the findings from the 53 clusters. Tables 3a and 3b present a summary of youth in the different risk categories in Mexico and Chile respectively while the full cluster results are presented in Tables 4a-4j. Annex B gives a brief description of each of the 53 clusters. A. Who Are the Youth at Risk in Mexico and Chile? Type III Risk Thirty-three percent of Mexican youth can be classified as type III risk, meaning that they are coping with the outcomes, or consequences, of their risky behavior, while 16.8% of Chilean youth fall into this risk category (Tables 3a and 3b). Considering separately the age- gender categories better characterizes type III risk youth in Mexico and Chile. Fourteen percent of Mexican youth aged 12 to 14 are in this risk category. They have almost all dropped out of school, resulting in an average education level of primary school. Few are involved in organized activities and a large proportion is working (49% of boys and 27% of girls). We cannot report on the sexual activity of these youth since they were not asked these types of questions. All youth who are illiterate are in this category (Tables 4a and 4b, column 1). Eight percent of Chileans age 15 to 17 and almost 40 percent of Mexicans in the same age range are classified as risk type III (Tables 3a and 3b). Almost all have dropped out of school prior to completing secondary education, many are inactive, most engage in risky sexual behavior and feel socially excluded. Among the women, a strong at-risk mother 13 group emerges, comprised of women who engaged in risky sexual behavior at a young age, left school, had their child, and largely remain unmarried.23 The males report unsafe sexual activity, low education level and idleness. They also report an early working age as well as an early age at onset of sexual activity (Tables 4c, 4g, 4h column 1; Table 4d columns 1-3). Almost half of Mexican youth (43%) and almost one-quarter of Chilean youth (22%) aged 18 to 24 can be classified as risk type III. These young people have had significantly more time to engage in risky behaviors and thus demonstrate a wider range of negative behaviors and outcomes. They break into two general categories: inactive youth and parents. The inactive males dropped out of school early and are not working ­ these are the stereotypical youth at-risk. Inactive females tend to be mothers who dropped out of school early and continue to engage in unsafe sexual activities. Among male parents, the fathers that are working typically dropped out of school prior to completing secondary education. Some but not all of these youth engage in risky sexual practices (Tables 4e, 4f, 4i, 4j).24 Several of the risk factors clustered with the groups described above, and protective factors were notably absent. The 12-14 year old Mexicans in this category come from the poorest households, tend to live in more rural areas, come from abusive households, had little parental control/influence on their behavior related to alcohol and smoking, had low access to healthcare and high social exclusion. The youth in these clusters report a low to average level of spirituality and church attendance (Tables 4a and 4b). Risk factors for the 15 to 17 year olds include low parental education, low level of connectedness, and non-positive family backgrounds. Thirty-three percent of the males in the Mexico type III risk clusters are from rural areas, but the Chilean type III clusters are not as rural with 25 to 27 percent living in rural areas25. Since Chile is much less rural overall than Mexico (13% compared to 25%), we can conclude that living in a rural area in Chile is more risky compared to living in a rural area in Mexico (Tables 4c, 4d, 4g, 4h). The 18 to 24 year olds in this age category come from disadvantaged backgrounds, with parents having below a secondary education on average. Twenty-three percent of the Chilean women in this category and age group live in rural areas, whereas about 19% of 23While the girls and boys in this cluster have similar situations, the data do not well capture fatherhood, so the most risky boy clusters will not report children, pregnancy or marriage, where the girls do. 24The males age 18-24 clusters that are classified as risk type III are largely defined by school and work variables, while the women's clusters are also defined by sexual behavior. This is largely driven by the data in two ways. First, the data about risky behavior sexual is largely targeted toward women. Second, the main risky behaviors that men engage in ­ violence and substance abuse ­ are not well captured in the data. 25Note: the tables for Mexico rural use 3 levels of rurality/urbanity: 1 = rural,, 0.5=from a small town, 0=urban. The information for a binary rural variable is reported above for comparison to Chile. 14 Chilean males in this age group and of this type live in rural areas. The "at-risk parents" are more rural than other groups (23% Chilean female cluster and 31% Mexican women are rural). The Chilean mothers that are early dropouts in this group report a high level of social exclusion, average family cohesion and are made up of 13% who self identify as indigenous. The "at-risk and inactive" clusters have the worst outcomes being idle (Tables 4e, 4f, 4i, 4j). A small group of males ­ 7.5% of Chileans and 11.4% of Mexicans ­ categorized as risk type III come from relatively good backgrounds. The Chileans tend to be in the 18 to 24 age range, have caring families and good behaviors in general (best sexual behaviors of all the clusters) and yet are idle (Table 4i, column 2). None are married and few are registered to vote. These idle youth did not drop out of secondary school and are the youngest cluster in this cohort, being under 20 on average. Perhaps this reflects the difficulty some youth have in finding a job to settle into when first graduating­ a closer look at the question that generated the work variable shows that 60% of these youth are looking for work. Overall, they are not very optimistic about the future, and do not feel very prepared for work. The Mexican men dropped out of school early, but most are working (99%). They were younger than most other clusters when they first started working and when they started their sexual activity and had children. Seventy-one percent are fathers and 99% are married (well above the other clusters), yet they are slightly more likely to have sexual relations with more than one person in the past year than they other clusters (Table 4e, column 3). Notably, this group is very small, suggesting that youth who have backgrounds with numerous protective factors can experience negative outcomes, but this is the exception; most youth who experience negative outcomes also have many risk factors in their lives. Figures 1a and 1b show the average behavior and factor levels for each cluster type. First considering Type III, we see the clusters of this type have lower averages for those variables that might be considered protective factors and high values for those variables that might be considered risk factors. No Risk At the other end of the spectrum is 20-45 percent of the youth population that has neither engaged in risky behaviors, nor demonstrated a high level of potentially risky factors. There are fewer youth in this category in the older age groups than in the younger age groups (Tables 3a and 3b). In addition to the low presence of risk factors, these youth have a higher level of protective factors and good behaviors than those with type II or III risk levels (Figures 1a and 1b). All of the 12-14 and 15-17 years old youth with type 0 risk are in 15 school, few are working, none have children, and all initiated sexual behavior at a later age - indeed most have not initiated sexual behavior - than those in groups classified as risk types I, II, or III. Among the 18-24 year old groups, many with risk type 0 are still in school, fewer are working than in the other risk type groups, and fewer have started families. Generally speaking, these youth can be considered more economically advantaged than other youth, while youth in this group who are not from wealthy households have strong emotional connections with their families and feel connected to adults who care about them. They tend to have fewer risk factors, in terms of household abuse or exposure to drugs (Tables 4a-4j, last columns in each table). Type I Risk Those with a higher risk than type 0 but much better than risk type III are the 9 to 20 percent of the Chilean and Mexican samples that have many risk factors, but they are not engaged in risky behaviors nor suffering negative consequences, i.e. they can be classified as type I risk (Tables 3a and 3b). Their behaviors are similar to those with no risk despite having a higher level of risk factors. In some cases this is perhaps because of the level of protective factors in their lives, such as connectedness with adults outside of the household, extracurricular activities or spirituality, and in others resiliency is likely playing a part while others are on their way to graduating to type II risk. One group of Chilean males aged 18 to 24 (9.3% of the cohort) is resilient and yet has poor protective factors to make up for the poor family cohesion with some substance abuse and abuse in the home. This group did not drop out of school early, half are still in school, yet about 20% are idle and 10% report victimization (Table 4i, column 5). Rural youth are over represented in some clusters, and underrepresented in others. Across age groups, there are fewer youth in this category in the older age groups than in the younger age groups as seen in Tables 3a and 3b. These youth are either too young to have begun engaging in risky behaviors, or they may be considered resilient. Some youth in this risk level (across all age groups, genders and countries) report a high level of social exclusion and can be considered loners. Those that report a high level of social exclusion and have good behaviors tend to report a higher level of connectedness with their families (better relationships, doing more activities with their parents, better cohesion, etc.), suggesting that being socially excluded from their peers could indeed be a protective factor in some cases. For others, though, the social exclusion may lead to deviant behaviors, as demonstrated by very high levels of social exclusion among older youth of risk type III. 16 The level of risk factors, protective factors and behaviors for this type of risk can be seen in Figures 1a and 1b for Mexico and Chile respectively. Behaviors are better than for types II and III risk and comparable to type 0 risk. Risk factors are worse than those with type 0 risk, but better than those with types II and III risk. Protective factors in Mexico are comparable to those with no risk; however those who are classified as type I risk in Chile have the fewest protective factors of any risk type, suggesting that this is a temporary phase with potentially worse to come. Type II Risk The most heterogeneous of the risk groups are the youth in clusters of risk type II who are 28% of the Chile sample and 25% of the Mexico sample. In Mexico, they are younger than the risk type III youth, with 33 percent of Mexicans age 12 to 14, 12 percent of Mexicans age 15 to 17, and 26.8 percent of Mexicans age 18 to 24 in this group. In contrast, about 20 percent of Chileans in the 15 to 17 age group and 33 percent in the 18 to 24 age group fit in this category. These youth are engaging in risky behaviors but are not suffering the consequences of their behaviors (Tables 3a and 3b). Figures 1a and 1b show that these youth have worse behaviors and outcomes compared to those with types 0 and I risk, however are not nearly as bad as those with type III. The types of behaviors vary amongst the different age groups and genders. The 12 to 14 year olds have a high level of early workers yet are still in school, in contrast to the most at-risk groups which are not in school. The boys falling into this category started work on average much earlier than in the other clusters, including the type III youth. For these youth, early work does not appear to influence their other behaviors but since they are younger than type III youth, there is a chance that some will "graduate" to risk type III. On the other hand, the girls in this category did not start work as early as the most at-risk cluster, and not as many of them are working (Tables 4a and 4b, columns 2). Among the 15-17 year olds, those with risk type II started sexual activity and work (more so for males) at a young age. They are still in school and are not parents, but their risky behaviors increase the likelihood that they will dropout or have children (Tables 4c, 4d, 4g, and 4h). Among the 18 to 24 year olds, a wide range in behavior for youth with type II risk is observed. Some have high levels of many risky activities, while others strongly display only one risky behavior (social exclusion or risky sex). A special group that breaks out can be identified as "resilient" youth who display type II risk, but have the luck, genetics, or 17 psychological make-up such that they have not suffered the consequences of their actions. For example, the resilient groups might have started work at an early age and yet managed to graduate from high school (Tables 4e, column 5; Table 4f column 5; Table 4j column 2). These youth tend to come from families with low parental education (though not as low as the most at-risk clusters), limited access to healthcare, and poor relationships with their parents. The 12 to 14 year olds have parents with low education and have poor relationships with their parents, but they have better relationships with their parents and a higher level of spirituality than those classified as risky type III. These youth may be on their way to the risk type III later in life or they may be resilient to the type I and II risk present in their lives. In the 15 to 17 age range, the different sub-groups have high levels of risk factors (low parental education, limited access to healthcare, and poor family cohesion) and medium levels of protective factors. The females tend to come from particularly difficult family backgrounds but have a high level of connectedness, often with an adult outside the family. The youth in the 18 to 24 age group tend to be more connected or resilient than those with type III risk, or have a significantly lower level of risk factors present (Tables 4a- 4j). B. Correlations among Risk Factors, Protective Factors, Risky Behaviors, and Outcomes Plotting risk or protective factors against behaviors and estimating the correlation confirms that young people with worse personal, micro, and macro influences in their lives are more at-risk than those who have positive influences. To keep this discussion short, we will discuss these relationships only for the 18 to 24 year old cohort in each country. The general trends described here are also observed when analyzing the younger cohorts as well as when grouping the clusters by each variation of age/gender groups.26 Good behaviors are positively correlated with protective factors in both Mexico and Chile as seen in Figures 2a and 2b respectively. The relationship is significant in Mexico for all age cohorts individually as well as when grouped together, however is not significant in Chile. The weaker results in Chile are potentially due to the lower number of available protective factors in its dataset. The analysis was also performed on youth aged 12 to 29 in 26These results are available from the authors upon request. While the relationships are not significantly different from zero in all cases due to the limited number of data points once disaggregating by gender and age, the relationships between factors and behaviors are clearer. Similarly, when grouping all clusters together for all ages and both genders, the results are not always interesting due to the wide difference in behaviors amongst age groups. 18 both countries as a robustness check27. Results were significantly different than 0 (at a 5% level) with this wider age range in both countries, suggesting the importance of age in some behaviors and outcomes. As suggested by the ecological framework, the opposite results are observed when considering the correlation between risk factors and positive behaviors (Figures 3a and 3b) for 18-24 year old youth. The relationships are significant in Mexico at a 10% level and not significant in Chile at the 10 percent level. The relationship is strengthened (1% level) in Mexico when grouping all clusters together rather than breaking up the different age groups. This is partially due to few risk factors in the Mexican data as compared to protective factors. The Chilean dataset, which is richer in risk factors than in protective, shows a negative and significant relationship at 1% between risk factors and positive behaviors in the 15 to 17 year old age group, but only at 15% in the 18 to 24 year olds28. C. Proxies for At-Risk Due to the nature of the data used in this analysis, we can use a large set of risk and protective factors to predict the level of risky behavior that will be displayed by youth. Some of our variables are easily observable characteristics and are regularly found in most household data sets, but most of these data are unique and not available for most countries and youth. Thus, it would be useful to identify a youth's risk level by a single or small set of easily observable characteristics. We test a set of observable variables to determine their usefulness in predicting risky behavior: poverty, living in a rural area, ethnicity, gender, and age. Poverty is a fairly good proxy to identify youth at-risk. Tables 3a and 3b summarize the information from the clusters and reveal that as the risk type rating of a group increases, the average level of parental education falls. Plotting the clusters against wealth proxies and estimating the slope of the line confirms this relationship.29 Figure 4a shows that in Mexico, the more at-risk clusters have lower parental education, and the relationship is statistically significant at the 1% level. The same result emerges with each Mexico age group as well as when merging the Mexico age groups together, with a negative and significant coefficient on 27Results are available upon request. 28The relationship was significant at 10% and negative for the 18 to 29 year old cohort. 29Since poverty level is an exogenous variable, we test these relationships for all poverty indicators: education level of household head, household ownership of luxury or durable goods, or monthly earnings of heads of households. Since there are no significant differences between the various proxies for poverty, we chose parental education level since it is a widely available and observable indicator. 19 poverty for all age groups. The coefficients are larger negative values for the younger age groups. Similarly, Chile's graph of poverty against positive behaviors in Figure 4b shows a statistically significant downward trend of behavior against poverty (proxied by parental education).30 This negative trend is significant for the younger cohort as well, with an even larger magnitude. While poverty is a reasonable candidate as an indicator in identifying at-risk groups,31 we cannot conclude that economic class is the sole source of clustering since a close look at the clusters shows that not everyone in the at-risk groups is poor. The figures in Annex C use the density of the various possible general levels of education of heads of households for each cluster and demonstrate that poverty is correlated with risk level, however each risk cluster has youth with parents of varying education levels. Furthermore, we do not suggest that poverty causes risky behavior, but, poverty, for whatever reason, can be used to identify ex ante which youth have a higher propensity for engaging in risky behaviors. Living in a rural area is a candidate for a proxy for risk; in both countries, rural youth are over 10 percentage points more likely to be in risk types I, II, and III than are urban youth, and ten percentage points less likely to be risk-free (Tables 3a and 3b). Figures 5a and 5b confirm this correlation, demonstrating that living in a rural area is negatively correlated with positive behaviors and outcomes at the 1% level. That said it is important to note that not all rural youth are at risk, and that most at-risk youth live in urban areas, since more of the general population lives in urban areas.32 Again, in Chile living in a rural area may be more risky than in Mexico, since the more at-risk clusters in Chile are more rural compared to the national average than Mexico. Ordered logistic regressions relating the risk level to poverty level and living in a rural area (when controlling for age and gender as well as when not doing so) demonstrate a large, significant and negative relationship for both variables as shown in Table 5. This 30Similarly Chile had two variables available to represent poverty: parent/caregiver education level and the economic class variable from the survey data (nse ­ nivel socio-economico). The "nse" variable was constructed by the Chilean National Institute of Youth using parental/caregiver education and occupation or an average of six durable goods in the event that the other information was not available. 31Indeed, Blakely et al. (2005) used income level to estimate relative risk levels. Their risk factors are: tobacco, drinking alcohol, not having access to safe water/sanitation, indoor air pollution exposure and obesity. 32In Mexico, the rural indicator was constructed to equal 1 if rural (< 2500 inhabitants), 0.5 if living in a small to medium sized town (2500 to 99,000) and 0 if living in a large urban setting. The rural indicator in Chile equals 1 if rural with more than 2000 and less than 5000 inhabitants and 0 if urban with more than 5000 inhabitants. 20 suggests that using both poverty as well as rurality might prove useful in identifying the at- risk youth. Ethnic identity is not a good proxy for youth at-risk, although indigenous youth are over-represented among clusters defined as risk type III (Table 3b). A closer look at the indigenous youth in the Chile sample shows that while indigenous youth have worse risk factors and behaviors, they have better protective factors.33 The more at risk clusters are about 5 percentage points more indigenous than the average, but the negative relationship between risk factors and positive behaviors is slight and insignificant as seen in Figure 6.34 We must remember, however, that this is a self-reported survey, with the youth indicating if they identify with a particular indigenous group.35 Overall levels of risky behavior increase with age, as seen in Tables 3a and 3b. The bubble graphs by age confirm this observed correlation.36 Older youth have lower levels of positive behaviors across the board. On the other hand the poverty level for the most at risk clusters is higher for the younger cohorts than the older, indicating that with age a wider variety of background characteristics are involved in risky behaviors. In Mexico, gender does not seem to matter (Table 3a) while in Chile women seem to be less at-risk (Table 3b). A closer look at the data reveals that male risky behavior is related to education and employment and females are more at risk for poor sexual behavior at younger ages. The females participating in risky behaviors at young ages tend to exhibit many risky behaviors, more than the males in the same cohort. Unfortunately, we were not able to look at violent or criminal behavior for this study, which is generally considered to be more common among men. In both Chile and Mexico, the most at-risk groups of females are more rural than males, despite the fact that similar percentages of males and females (across all age ranges) are rural. So being female and living in a rural area is more risky than being male and living in a rural area37. Finally, we see a larger percentage of youth at risk in Mexico than in Chile, perhaps reflecting overall macroeconomic growth or government policy. 33Further details available upon request. 34Youth 15 to 17 years of age have a similar plot which is available upon request. 35See Hall and Patrinos (2006) for documentation of the limitations of using self-reported ethnicity. 36This is seen with both the Chile and the Mexico clusters by comparing the different age cohorts. We do not present the graphs here but they are available upon request. 37Approximately 13% of females and males live in rural areas in Chile and 24.5% in Mexico. Of those with type III risk, 20% of males and 23.6% of females in Chile and 29.7% of males and 33.7% of females in Mexico live in rural areas. 21 VI. Conclusion Based on unique youth surveys in Chile and Mexico, we find that more than half of young people in these two countries can be considered at-risk. In Mexico, nearly one-third of people age 12-24 are suffering the consequences of negative youth behaviors ­ adolescent mothers, early school dropout, not working ­ while 17 percent of Chilean youth are in this situation. They also come from the poorest families and have the fewest social bonds. Another one-quarter of Mexican and Chilean youth are engaging in negative behaviors and on their way to the worst-off category. A younger group in Mexico (20 percent) and Chile (8 percent) are not engaging in risky behaviors, but they have factor in their lives that suggest that they may be graduating to these more harmful groups before long since they also lack the social supports and mental health that the no-risk group boasts. These findings confirm the U.S. literature that negative factors in a young person's life is highly correlated with negative behaviors. However, further analysis, preferably with panel data, is needed to understand the transition more clearly. Some factors repeatedly arise in the best-off or worst-off groups. Positive relationship with the family is a recurring protective factor across the clusters. The clusters show that youth who live(d) with both parents have a lower incidence of all risky behaviors. Those youth who feel connected to a parent, i.e. those who feel that they can relate to a parent, the parent cares for them, they can depend on the parent, etc, also have lower risky behavior and negative outcomes than youth who do not feel connected. Connection with non-family members can partially compensate for absent parental connection, as shown by youth who are connected to non-family members having less risky behavior than those who do not have connections with anyone. Conversely, family abuse, substance abuse and lack of family cohesion cluster with victimization in Chile, while non-positive feedback from parents is correlated with risky behaviors in Mexico. Positive institutional factors are also correlated with positive behaviors. School quality matters in Chile, and an individual's relationship with his/her community (trust, feeling of school quality) is positively correlated with voter registration. Low spirituality/church attendance clusters loosely with the risk type III groups. Surprisingly, social exclusion is rarely correlated with negative behaviors. Instead, the "loner" youth usually showed up as sub-groups in the less at-risk clusters. Negative behaviors (or outcomes) cluster together. An early age of starting work clusters loosely with an early age at onset of sexual activity for boys and girls. Other risky behaviors by females cluster with risky sexual behavior while the pattern is weaker for males: 22 For those 12 to 17 and for all females, inactivity and early school drop cluster, which is not the case for the older males. For women 18 to 24 years of age, inactivity and marriage cluster together with a high incidence of having a child, however in this age range this is not risky behavior. There is no simple answer as to which variables can be used to proxy being at-risk, however poverty is the best option of those available. Poor youth are more at risk on average. Rural youth are more at risk on average than urban youth. When considering prevention programs to target at risk youth, it is important to remember that risk profiles vary with youth development. Young youth not exhibiting any risky behavior are not out of danger. Therefore age appropriate programs targeting those with type I risk could potentially prevent risky behaviors later in life. In addition, intervention plans should take gender into account. Gender appears to matter more so with regards to sexual health as younger at-risk females tend to participate in all risky behaviors available in our dataset, while young males exhibit some risky behaviors but not all. The public health literature suggests that perhaps violent behavior is more important for boys however this could not be tested with these data used in this study. 23 Table 1 ­ Construction of Variables Risk Type If variable appears Used in I, II, III in the data set the Chile Mexico cluster analysis Behaviors/outcomes Not idle ­ either in school nor working II, III X X X No early school dropout (i.e. completed secondary school) II X X X Literate (can read and/or write a message) III X X In school II, III X X Years of education completed I, III X X Working (Not Workingc) II, III X X Older age when started working II, III X X X Safe sex (among the sexually active) II X X Low number of sexual partners in the past year II Xa Older age at onset of sexual activity II X Xa X Older age at first pregnancy/child III X Xa X Has at least one child (Does not have at least one childc) III X X Married (Not Marriedc) I, III X X X Participate in extracurricular activities II X X X Registered to vote, planning on voting in the next election, III X Xb X desires to vote when of age (if under 18 years old) Has not been a victim of a crime (proxy for criminal activity) II X X Attitude towards drugs (respondent can justify using drugs) II Xa Attitude towards alcohol (respondent can justify getting drunk) II Xa Protective Factors Trust in governmental institutions I Xd Xa,e X Trust in community institutions I Xf Xa,g X Connected (whether youth reaches out ­ for talk or help- to I X X X someone when they have problems) Living with both, one or no parents I X X Positive relationship with father I Xh Xi X Positive relationship with mother I Xj Xk X Connected with an adult other than parents I X X Communication with parents (talk to parents when facing a I X personal problem) Church attendance I X X X Spiritual influence in beliefs, opinions and attitudes I X School qualityl I X X Feeling optimistic about future work I X Feeling prepared for future employment I X Sense of wellbeing (level of happiness reported by the youth) I Xa Risk Factors Have felt discriminated against I X Parental education ­ this is a good proxy for poverty I X X X Household ownership of durable/luxury goodsm I X Monthly earnings of heads of household I X Limited access to healthcare I X X 24 Rural residence (versus urban) I X X Social exclusion I, III X X X Poor family cohesion I X X Abuse in the home I X X Substance abuse in the home I X X Level of perceived violence in the neighborhood I X X Indigenous (self-identifying with an indigenous group) I X Parental influence with regards to smoking and alcoholn I X X Parental response to misbehavioro I X Parental response to good behaviorp I X afor respondents age 15 or older b for respondents age 18 or older cfor respondents under age 18 d indicates level of confidence in government, congress, city government, political parties, judicial system eindicates level of confidence in politicians, judges, the police, and the military flevel of confidence in hospitals, the Catholic Church, schools, universities, and family gindicates level of confidence in teachers, doctors, shop owners, union leaders and priests h quality of relationship with father on various attributes (communication, demonstration of love or affection, understanding and help with problems, respect for private life of youth, the time spend with father) ivariety of topics that the youth communicates with the father about (school, politics, religion, sexual relations, work, and other topics) jindicates quality of relationship with mother on various attributes (communication, demonstration of love or affection, understanding and help with problems, respect for private life of youth, the time spend with mother) k variety of topics that the youth communicates with the mother about (school, politics, religion, sexual relations, work, and other topics) lrank of the overall quality of the youth's current/past school as reported by youth (physical building, scholastic materials, teachers preparation, content of courses and teachers assistance) m includes radio recorder, CD burner, TV, cable, VCR, game console, telephone, computer, internet, car/truck/van - could be used instead of household education level to indicate economic class nindicates level of control parents attempt to control children's behaviors (do they forbid smoking/drinking, grant periodic permission, allow the child to make his/her own decision) ohow parents respond when child bothers/angers them (0=by talking with their child, 0.5 = punishing, 1 = beating/hitting, insulting, accusation in front of others, stop talking) p indicates frequency with which parents use positive feedback (words of encouragement, hug/kiss, give a gift, concede to something) when child does something good/correct (0 = always, 0.5 = sometimes, 1 = never) 25 Table 2 ­ Descriptive Statistics Mexico Chile All 12 to 15 to 18 to All 15 to 18 to Actual percentages ages 14 17 24 ages 17 24 Average age 17.19 13.0 16.0 20.8 18.9 15.95 20.8 Percent female 53.5 50.65 52.9 55.9 53.0 50.5 54.6 Percent indigenous - - - - 10.8 11.5 10.4 Percent rural 24.7 28.1 26.0 21.6 13.2 13.3 13.1 Behaviors and Outcomes Share not completing high school 29.9 10.0 33.0 42.3 14.2 8.4 17.8 Share in school 45.3 85.9 61.8 27.5 60.5 91.4 41.0 Share idle/inactive 20.3 8.9 17.7 29.9 27.3 6.7 35.9 Share working 41.9 17.2 35.1 35.1 21.4 4.7 32.0 Share having sex 52.1 - 13.3 56.0 59.1 26.9 79.4 Share of the sexually active using protection 50.6 - 54.0 52.5 61.9 59.9 62.3 Share reporting at least 1 child 19.4 - 2.5 28.9 18.6 3.9 27.9 Share married 15.7 0.4 4.1 32.9 4.8 0.1 7.7 Risk and Protective Factors Share reporting physical or psychological abuse - - - - 6.3 6.2 6.4 in home Share reporting problems arising from substance - - - - 8.5 7.0 9.5 abuse in home Share without access to medical services 49.4 50.2 50.7 48.0 - - - Share attending church weekly 9.3 11.3 10.3 7.3 21.6 20.9 18.1 Share attending church at least once in the past 66.3 72.0 65.9 62.5 - - - month Share believing in God - - - - 94.8 95.1 94.6 Share optimistic about the future - - - - 87.7 88.2 87.3 Share reporting being happy (sense of well- 98.2 - 98.0 98.4 - - - being) Share reporting social exclusion 53.2 66.8 53.9 43.1 17.7 10.1 22.5 Share with parents who have a primary degree or 56.1 56.3 58.0 55.0 38.1 38.8 37.7 less Share reporting they have felt discriminated - - - - 54.7 54.2 55.1 against 26 Table 3a ­ Mexico Cluster analysis overview Type III Type II Type I None Percent of total sample 33.1 25.1 20.2 21.6 Percent of 12-14 year olds 14.1 33.1 24.3 28.5 Percent of 15-17 year olds 38.9 11.9 27.5 21.7 Percent of 18-24 year olds 43.4 26.8 13.1 16.7 Percent of males 31.9 23.2 22.2 22.7 Percent of females 34.2 26.7 18.4 20.7 Average Parental Primary graduate, Primary graduate, Secondary Secondary Education (poverty proxy) some secondary some secondary graduate graduate Percent of rural 42.9 24.4 18.1 14.6 Percent of urban 29.9 25.3 20.8 23.9 Table 3b ­ Chile Cluster analysis overview Type III Type II Type I None Percent of total sample 16.8 28.0 8.7 46.5 Percent of 15-17 year olds 8.3 20.4 15.9 55.5 Percent of 18-24 year olds 22.2 32.8 4.2 40.8 Percent of males 23.7 19.1 13.2 44.1 Percent of females 10.8 4.7 35.9 10.8 Average Parental Primary Graduate Some Secondary Some Secondary Secondary Education (poverty proxy) Education Education Graduate Percent of rural 27.2 29.5 7.0 36.3 Percent of urban 15.3 27.7 9.0 48.0 Percent of indigenous 21.7 26.3 8.7 43.3 Percent of non-indigenous 16.3 28.2 8.7 46.9 27 Table 4a: Cluster analysis results: Mexican Males Aged 12 to 14 Cluster name at-risk very early advantaged advantaged workers loners youth age 13.25 13.07 12.87 13.05 Behaviors Average 0.41 0.69 0.83 0.77 older age when first started working 0.42 0.11 1.00 0.62 not idle 0.49 1.00 1.00 1.00 no early school dropout 0.01 1.00 1.00 1.00 literacy 0.91 1.00 1.00 1.00 not married 0.99 1.00 1.00 1.00 participate in activities 0.14 0.29 0.17 0.23 not working 0.51 0.58 0.99 0.81 in school 0.01 1.00 1.00 1.00 years of education completed 0.20 0.28 0.27 0.28 Protective Factors Average 0.54 0.58 0.60 0.58 relationship with father 0.37 0.42 0.40 0.40 relationship with mother 0.42 0.48 0.46 0.46 connected 0.96 0.84 1.00 1.00 live with both parents 0.87 0.90 0.93 0.90 church attendance 0.36 0.41 0.39 0.39 school quality 0.75 0.81 0.83 0.82 communication with parents 0.49 0.52 0.55 0.51 spiritual influence 0.15 0.22 0.15 0.17 Risk Factors Average 0.57 0.57 0.54 0.41 parental influence (alcohol & smoking) 0.44 0.38 0.35 0.37 social exclusion 0.58 1.00 1.00 0.00 limited access to healthcare 0.63 0.52 0.45 0.47 parental response to misbehavior 0.38 0.31 0.29 0.32 parental response to good behavior 0.58 0.52 0.51 0.53 rural 0.55 0.49 0.45 0.43 parental education 0.80 0.70 0.67 0.67 household ownership of goods 0.84 0.79 0.79 0.77 monthly earnings household heads 0.85 0.80 0.78 0.78 n 758 1193 1893 1989 Risk Type III II I - Percent of sample 13.0% 20.5% 32.5% 34.1% Notes: The highlighted rows indicate that the variables were not included in the clustering exercise. They are exogenous veriables. 28 Table 4b: Cluster analysis results: Mexican Females Aged 12 to 14 Cluster name at-risk early workers advantaged advantaged loners youth Age 13.33 12.98 12.97 13.13 Behaviors Average 0.43 0.79 0.83 0.80 older age when first started working 0.66 0.74 1.00 0.77 not idle 0.28 1.00 1.00 1.00 No early school dropout 0.01 1.00 1.00 1.00 literacy 0.92 1.00 1.00 1.00 not married 0.97 1.00 0.99 1.00 participate in activities 0.13 0.24 0.24 0.25 not working 0.73 0.87 0.99 0.89 in school 0.01 0.99 1.00 1.00 years of education completed 0.20 0.27 0.29 0.29 Protective Factors Average 0.56 0.59 0.61 0.59 relationship with father 0.36 0.39 0.42 0.40 relationship with mother 0.45 0.50 0.52 0.51 Connected 0.95 0.93 1.00 1.00 live with both parents 0.86 0.88 0.93 0.90 church attendance 0.45 0.44 0.42 0.45 school quality 0.77 0.82 0.81 0.81 communication with parents 0.45 0.50 0.55 0.48 spiritual influence 0.18 0.17 0.17 0.18 Risk Factors Average 0.66 0.65 0.48 0.48 parental influence (alcohol & smoking) 0.40 0.48 0.06 0.34 social exclusion 0.73 1.00 1.00 0.00 limited access to healthcare 0.64 0.69 0.00 0.43 parental response to misbehavior 0.30 0.27 0.26 0.27 parental response to good behavior 0.59 0.51 0.47 0.51 Rural 0.59 0.49 0.35 0.40 parental education 0.81 0.71 0.60 0.65 household ownership of goods 0.86 0.81 0.76 0.76 monthly earnings household heads 0.87 0.81 0.74 0.75 N 908 2723 982 1376 Risk Type III II I - Percent of sample 15.2% 45.5% 16.4% 23.0% Notes: The highlighted rows indicate that the variables were not included in the clustering exercise. They are exogenous veriables. 29 Table 4c: Cluster analysis results: Mexican Males Aged 15 to 17 Cluster name at-risk becoming at loners advantaged risk youth age 16.16 15.89 15.88 15.97 Behaviors Average 0.64 0.84 0.85 0.84 older age when first started working 0.25 0.41 0.52 0.48 not idle 0.74 1.00 1.00 1.00 no early school dropout 0.06 1.00 1.00 1.00 literacy 0.97 1.00 1.00 1.00 older age at onset of sexual activity 0.85 0.92 0.90 0.88 older age at first pregnancy 0.98 1.00 1.00 1.00 not married 0.97 1.00 1.00 1.00 participate in activities 0.17 0.27 0.27 0.32 not working 0.30 0.63 0.73 0.73 in school 0.06 0.99 0.99 1.00 years of education completed 0.28 0.39 0.42 0.42 safe sex 0.85 0.93 0.91 0.90 low number of sexual partners in past year 0.96 0.98 0.98 0.97 does not have a child 0.99 1.00 1.00 1.00 attitude towards drugs 0.97 0.97 0.96 0.95 attitude towards alcohol 0.88 0.89 0.91 0.85 Protective Factors Average 0.54 0.58 0.59 0.58 Connected 0.91 1.00 1.00 1.00 live with both parents 0.84 0.88 0.89 0.88 relationship with father 0.38 0.42 0.44 0.43 relationship with mother 0.44 0.47 0.50 0.49 trust in governmental institutions 0.35 0.36 0.36 0.38 trust in community institutions 0.58 0.60 0.60 0.61 church attendance 0.32 0.35 0.35 0.32 school quality 0.78 0.81 0.81 0.82 communication with parents 0.44 0.48 0.51 0.47 spiritual influence 0.19 0.19 0.18 0.18 sense of wellbeing 0.82 0.85 0.89 0.87 Risk Factors Average 0.54 0.57 0.45 0.29 parental influence (alcohol & smoking) 0.54 0.46 0.42 0.45 social exclusion 0.46 0.48 1.00 0.00 limited access to healthcare 0.61 1.00 0.04 0.00 parental response to misbehavior 0.29 0.27 0.21 0.24 parental response to good behavior 0.60 0.55 0.50 0.53 rural 0.51 0.49 0.35 0.29 parental education 0.78 0.72 0.59 0.56 household ownership of goods 0.82 0.78 0.73 0.69 monthly earnings household heads 0.85 0.82 0.74 0.70 n 1765 1120 681 869 Risk Type III II I - Percent 39.8% 25.3% 15.4% 19.6% Notes: The highlighted rows indicate that the variables were not included in the clustering exercise. They are exogenous veriables. 30 Table 4d: Cluster analysis results: Mexican Females Aged 15 to 17 Cluster names early at-risk at-risk resilient advantaged advantaged workers" mothers idle loners loners youth & wives dropouts age 16.20 16.42 16.07 15.85 15.85 15.92 Behaviors Average 0.66 0.46 0.68 0.86 0.88 0.88 older age when started working 0.26 0.43 0.65 0.62 0.71 0.67 not idle 1.00 0.21 0.02 1.00 1.00 1.00 no early school dropout 0.00 0.07 0.00 0.96 0.99 1.00 literacy 1.00 0.96 0.95 1.00 1.00 1.00 older age at onset of sexual 0.97 0.29 0.97 0.99 0.98 0.97 activity older age at first pregnancy 1.00 0.43 0.99 1.00 1.00 1.00 not married 0.99 0.34 0.99 1.00 0.99 1.00 participate in activities 0.16 0.07 0.15 0.25 0.27 0.31 not working 0.00 0.83 0.98 0.77 0.86 0.83 in school 0.00 0.06 0.00 0.95 0.99 1.00 years of education completed 0.27 0.27 0.26 0.38 0.42 0.42 safe sex 0.96 0.19 0.96 0.99 0.98 0.97 low number of sexual partners in 0.99 0.89 0.99 1.00 1.00 0.99 past year does not have a child 1.00 0.46 0.99 1.00 1.00 1.00 attitude towards drugs 0.98 0.97 0.97 0.97 0.97 0.96 attitude towards alcohol 0.94 0.94 0.94 0.95 0.93 0.89 Protective Factors Average 0.57 0.49 0.57 0.57 0.62 0.60 connected 1.00 0.85 0.96 0.82 1.00 1.00 live with both parents 0.85 0.27 0.87 0.85 0.91 0.86 relationship with father 0.43 0.41 0.45 0.48 0.52 0.45 relationship with mother 0.39 0.35 0.36 0.41 0.44 0.42 trust in governmental institutions 0.35 0.32 0.34 0.33 0.36 0.37 trust in community institutions 0.59 0.57 0.58 0.58 0.60 0.62 church attendance 0.20 0.16 0.18 0.22 0.20 0.21 school quality 0.82 0.79 0.78 0.82 0.82 0.80 communication with parents 0.51 0.46 0.46 0.53 0.57 0.54 spiritual influence 0.42 0.31 0.45 0.45 0.43 0.42 sense of wellbeing 0.81 0.85 0.79 0.82 0.89 0.86 Risk Factors Average 0.54 0.53 0.57 0.62 0.43 0.35 parental influence (alcohol & 0.43 0.43 0.41 0.40 0.38 0.39 smoking) social exclusion 0.60 0.39 0.68 1.00 1.00 0.00 limited access to healthcare 0.62 0.63 0.64 0.92 0.01 0.39 parental response to misbehavior 0.26 0.38 0.28 0.25 0.22 0.24 parental response to good 0.59 0.58 0.58 0.52 0.47 0.49 behavior rural 0.48 0.50 0.61 0.50 0.33 0.32 parental education 0.78 0.77 0.80 0.72 0.60 0.60 household ownership of goods 0.83 0.84 0.84 0.80 0.74 0.72 monthly earnings household heads 0.85 0.85 0.86 0.83 0.74 0.73 n 603 363 932 957 947 1178 Risk Type III III III I I - Percent 12.1% 7.3% 18.7% 19.2% 19.0% 23.7% Notes: The highlighted rows indicate that the variables were not included in the clustering exercise. They are exogenous veriables. 31 Table 4e: Cluster analysis results: Mexican Males Aged 18 to 24 Cluster name idle drinkers working loners resilient advantaged dads loners students age 20.58 19.60 22.28 21.19 20.96 20.15 Behaviors Average 0.47 0.56 0.64 0.58 0.63 0.69 older age when started working 0.42 0.30 0.30 0.32 0.40 0.46 not idle 0.00 0.98 1.00 1.00 1.00 1.00 no early school dropout 0.27 0.51 0.11 0.00 0.25 1.00 literacy 0.99 0.90 1.00 1.00 1.00 1.00 older age at onset of sexual activity 0.66 0.69 0.50 0.62 0.71 0.68 older age at first 0.91 0.92 0.69 0.92 0.95 0.98 pregnancy/parenthood married 0.14 0.14 0.99 0.05 0.06 0.01 participate in activities 0.19 0.18 0.18 0.20 0.24 0.31 registered to vote 0.95 0.95 0.98 0.96 0.96 0.93 working 0.00 0.81 0.99 0.91 0.78 0.56 in school 0.00 0.30 0.07 0.20 0.38 0.74 years of education completed 0.36 0.34 0.37 0.35 0.44 0.69 safe sex 0.57 0.61 0.22 0.51 0.63 0.64 low number of sexual partners in 0.89 0.90 0.87 0.87 0.90 0.89 past year has a child 0.16 0.14 0.71 0.15 0.11 0.03 attitude towards drugs 0.82 0.84 0.86 0.82 0.85 0.77 attitude towards alcohol 0.73 0.06 1.00 1.00 1.00 1.00 Protective Factors Average 0.54 0.53 0.53 0.55 0.55 0.59 relationship with father 0.46 0.45 0.48 0.47 0.49 0.50 relationship with mother 0.40 0.40 0.43 0.42 0.43 0.48 connected 0.96 0.95 0.99 1.00 0.81 1.00 live with both parents 0.75 0.73 0.34 0.73 0.79 0.83 trust in governmental institutions 0.34 0.36 0.35 0.34 0.38 0.36 trust in community institutions 0.59 0.60 0.62 0.61 0.63 0.64 church attendance 0.30 0.28 0.31 0.30 0.31 0.31 school quality 0.79 0.76 0.78 0.80 0.80 0.81 communication with parents 0.47 0.45 0.48 0.48 0.49 0.53 spiritual influence 0.19 0.20 0.19 0.17 0.20 0.19 sense of wellbeing 0.82 0.83 0.91 0.85 0.84 0.89 Risk Factors Average 0.58 0.56 0.50 0.51 0.60 0.45 parental influence (alcohol & 0.66 0.66 0.68 0.74 0.73 0.72 smoking) social exclusion 0.37 0.36 0.10 0.00 1.00 0.16 limited access to healthcare 0.61 0.52 0.38 0.46 0.40 0.44 parental response to misbehavior 0.28 0.29 0.31 0.29 0.24 0.22 parental response to good behavior 0.59 0.60 0.59 0.60 0.56 0.54 rural 0.48 0.44 0.38 0.36 0.38 0.24 parental education 0.74 0.73 0.66 0.73 0.69 0.54 household ownership of goods 0.78 0.80 0.80 0.76 0.75 0.66 monthly earnings household heads 0.84 0.83 0.80 0.82 0.81 0.71 n 743 1511 837 1778 1335 1142 Risk Type III III III II I None Percent 10.1% 20.6% 11.4% 24.2% 18.2% 15.5% Notes: The highlighted rows indicate that the variables were not included in the clustering exercise. They are exogenous veriables. 32 Table 4f: Cluster analysis results: Mexican Females Aged 18 to 24 Cluster name at risk, at risk, poor somewhat resilient advantaged idle idle, and working advantaged poor students married excluded early married mothers married dropouts moms mothers age 21.35 21.09 21.27 21.48 19.02 20.27 Behaviors Average 0.50 0.51 0.63 0.63 0.62 0.72 older age when started working 0.53 0.56 0.47 0.46 0.48 0.63 not idle 0.00 0.07 1.00 1.00 0.95 0.89 no early school dropout 0.07 0.19 0.00 0.10 0.62 1.00 literacy 0.99 0.92 1.00 0.98 1.00 1.00 older age at onset of sexual activity 0.58 0.67 0.85 0.70 0.92 0.91 older age at first 0.60 0.68 0.87 0.76 0.97 0.96 pregnancy/parenthood married 0.63 0.40 0.12 0.45 0.03 0.05 participate in activities 0.11 0.11 0.18 0.18 0.22 0.29 registered to vote 0.98 0.98 0.97 0.98 0.97 0.97 working 0.00 0.05 0.83 0.82 0.60 0.40 in school 0.00 0.02 0.29 0.26 0.48 0.69 years of education completed 0.29 0.29 0.35 0.39 0.40 0.71 safe sex 0.36 0.50 0.75 0.54 0.88 0.86 low number of sexual partners in 0.90 0.93 0.96 0.93 0.98 0.98 past year has a child 0.68 0.56 0.23 0.40 0.05 0.06 attitude towards drugs 0.96 0.95 0.93 0.90 0.92 0.88 attitude towards alcohol 0.76 0.73 1.00 0.93 0.00 0.99 Protective Factors Average 0.53 0.51 0.57 0.56 0.58 0.61 relationship with father 0.43 0.43 0.46 0.45 0.46 0.51 relationship with mother 0.37 0.36 0.41 0.41 0.41 0.48 connected 1.00 0.64 0.85 1.00 1.00 0.95 live with both parents 0.26 0.52 0.74 0.47 0.80 0.80 trust in governmental institutions 0.35 0.35 0.35 0.35 0.34 0.36 trust in community institutions 0.61 0.61 0.62 0.62 0.61 0.63 church attendance 0.37 0.39 0.39 0.36 0.41 0.39 school quality 0.80 0.76 0.81 0.81 0.80 0.82 communication with parents 0.49 0.48 0.54 0.55 0.54 0.60 spiritual influence 0.18 0.20 0.21 0.22 0.22 0.25 sense of wellbeing 0.89 0.82 0.82 0.90 0.82 0.89 Risk Factors Average 0.51 0.65 0.59 0.42 0.54 0.46 parental influence (alcohol & 0.47 0.49 0.55 0.53 0.50 0.59 smoking) social exclusion 0.00 0.97 0.80 0.01 0.52 0.42 limited access to healthcare 0.51 0.65 0.57 0.14 0.54 0.36 parental response to misbehavior 0.36 0.33 0.28 0.30 0.26 0.21 parental response to good behavior 0.58 0.59 0.55 0.55 0.54 0.49 rural 0.45 0.54 0.40 0.26 0.40 0.24 parental education 0.71 0.75 0.74 0.63 0.72 0.55 household ownership of goods 0.82 0.83 0.79 0.76 0.78 0.68 monthly earnings household heads 0.84 0.86 0.84 0.75 0.82 0.72 n 1810 2324 1693 999 850 1644 Risk Type III III II II I - Percent 19.4% 24.9% 18.2% 10.7% 9.1% 17.6% 33 Notes: The highlighted rows indicate that the variables were not included in the clustering exercise. They are exogenous veriables. Table 4g: Cluster analysis results: Chilean Males Aged 15 to 17 Cluster name at-risk risky doing ok loners advantaged advantaged sex youth - some youth sexual activity age 16.31 16.01 15.90 15.95 15.78 15.72 Behaviors/Consequences Average 0.53 0.77 0.82 0.78 0.84 0.86 older age when first working 0.39 0.52 0.67 0.71 0.74 0.97 not idle 0.36 1.00 1.00 1.00 1.00 1.00 no early school dropout 0.00 1.00 1.00 0.98 1.00 1.00 older age at onset of sexual activity 0.60 0.63 0.83 0.73 0.86 0.98 older age at first pregnancy/child 0.97 0.99 1.00 0.98 1.00 1.00 not married 1.00 1.00 1.00 1.00 1.00 1.00 Participate in activities 0.09 0.16 0.18 0.11 0.23 0.14 not a victim 0.92 0.91 0.91 0.91 0.93 0.94 not working 0.64 0.92 0.95 0.92 0.98 0.99 in school 0.00 1.00 1.00 0.98 0.99 1.00 years of education completed 0.31 0.46 0.45 0.46 0.45 0.45 safe sex 0.58 0.62 0.82 0.72 0.87 0.98 does not have a child 0.96 0.99 1.00 0.97 1.00 0.99 desires to vote 0.53 0.55 0.63 0.44 0.64 0.59 Protective Factors Average 0.59 0.62 0.75 0.50 0.76 0.66 relationship with father 0.65 0.62 0.70 0.70 0.92 0.86 relationship with mother 0.85 0.88 0.79 0.86 0.95 0.93 Connected to other adults 0.05 0.02 0.93 0.00 0.00 0.02 Connected 0.72 0.96 1.00 0.00 1.00 1.00 trust in government 0.31 0.33 0.37 0.28 0.44 0.37 trust in community 0.74 0.79 0.83 0.74 0.84 0.81 Spirituality 0.26 0.21 0.34 0.23 0.97 0.22 Risk Factors Average 0.35 0.31 0.27 0.26 0.25 0.24 poor family cohesion 0.38 0.38 0.28 0.35 0.19 0.26 abuse in the home 0.09 0.09 0.00 0.01 0.00 0.01 substance abuse in the home 0.08 0.19 0.00 0.00 0.00 0.00 social exclusion 0.13 0.01 0.00 0.11 0.00 0.17 community violence 0.15 0.21 0.25 0.18 0.29 0.19 felt discriminated against 0.61 0.63 0.51 0.50 0.46 0.48 live in a rural area 0.27 0.13 0.19 0.08 0.13 0.11 indigenous 0.20 0.12 0.13 0.10 0.17 0.08 low economic class (nse) 0.85 0.71 0.74 0.71 0.67 0.63 low parental education 0.75 0.59 0.61 0.59 0.56 0.50 N 75 359 88 105 126 265 Risk Level III II I I - - Percent of sample 7.4% 35.3% 8.6% 10.3% 12.4% 26.0% Notes: The highlighted rows indicate that the variables were not included in the clustering exercise. They are exogenous veriables. 34 Table 4h: Cluster analysis results: Chilean Females Aged 15 to 17 Cluster name at-risk high risk but loners advantaged connected connected youth age 16.26 16.15 15.91 15.94 16.00 Behaviors/Consequences Average 0.44 0.79 0.81 0.83 0.82 older age when started working 0.59 0.78 0.84 0.86 0.88 not idle 0.07 1.00 1.00 1.00 0.99 no early school dropout 0.01 1.00 1.00 1.00 1.00 older age at onset of sexual activity 0.37 0.67 0.84 0.89 0.87 older age at first parenthood 0.61 0.99 0.96 0.99 0.99 not married 0.98 1.00 1.00 1.00 1.00 participate in activities 0.02 0.12 0.08 0.11 0.11 not a victim 0.92 0.91 0.94 0.93 0.92 not working 0.93 0.93 0.99 0.99 0.98 in school 0.00 1.00 1.00 1.00 0.99 years of education completed 0.31 0.45 0.47 0.46 0.47 safe sex 0.28 0.63 0.82 0.88 0.84 does not have a child 0.51 0.98 0.95 0.98 0.99 desires to vote 0.62 0.55 0.50 0.55 0.49 Protective Factors Average 0.61 0.60 0.55 0.66 0.74 relationship with father 0.59 0.55 0.57 0.75 0.71 relationship with mother 0.75 0.71 0.85 0.91 0.86 connected to other adults 0.12 0.20 0.02 0.00 1.00 connected 0.84 1.00 0.41 1.00 1.00 trust in government 0.38 0.24 0.30 0.31 0.29 trust in community 0.79 0.70 0.70 0.81 0.77 spirituality 0.28 0.36 0.32 0.42 0.42 Risk Factors Average 0.39 0.48 0.34 0.27 0.29 poor family cohesion 0.39 0.69 0.39 0.33 0.38 abuse in the home 0.14 1.00 0.05 0.00 0.00 substance abuse in the home 0.13 0.20 0.05 0.05 0.06 social exclusion 0.55 0.10 0.53 0.00 0.07 community violence 0.25 0.29 0.34 0.31 0.33 felt discriminated against 0.52 0.82 0.55 0.51 0.58 live in a rural area 0.25 0.18 0.08 0.12 0.10 indigenous 0.11 0.20 0.12 0.10 0.08 low economic class (nse) 0.85 0.73 0.73 0.69 0.71 low parental education 0.72 0.59 0.59 0.56 0.58 n 95 60 133 661 90 Risk Level III II I - - Percent of sample 9.1% 5.8% 12.8% 63.6% 8.7% Notes: The highlighted rows indicate that the variables were not included in the clustering exercise. They are exogenous veriables. 35 Table 4i: Cluster analysis results: Chilean Males Aged 18 to 24 Cluster name idle idle high working connected not well advantaged advantaged dropout school (fathers) protected students (dads) graduates - socially excluded age 20.99 19.98 21.02 20.65 20.43 20.41 21.72 Behaviors/Consequences 0.37 0.37 0.48 0.48 0.48 0.51 0.56 Average older age when started working 0.38 0.60 0.48 0.49 0.53 0.57 0.56 not idle 0.62 0.00 0.80 0.79 0.79 0.93 0.81 no early school dropout 0.28 1.00 0.80 0.85 0.99 0.98 0.93 older age at onset of sexual activity 0.41 0.54 0.53 0.48 0.47 0.51 0.48 older age at first parenthood 0.90 0.97 0.90 0.96 0.96 0.95 0.98 married 0.00 0.00 0.08 0.04 0.02 0.05 0.00 participate in activities 0.08 0.11 0.07 0.16 0.14 0.12 0.12 registered to vote 0.02 0.01 0.14 0.17 0.01 0.01 1.00 not victim 0.93 0.92 0.92 0.90 0.90 0.91 0.91 working 0.53 0.00 0.57 0.42 0.38 0.39 0.38 in school 0.13 0.00 0.28 0.44 0.52 0.64 0.56 years of education completed 0.43 0.61 0.55 0.61 0.62 0.65 0.68 safe sex 0.22 0.37 0.35 0.33 0.30 0.35 0.32 has a child 0.22 0.06 0.25 0.12 0.09 0.13 0.06 Protective Factors Average 0.58 0.66 0.62 0.78 0.50 0.63 0.66 relationship with father 0.56 0.85 0.69 0.74 0.61 0.68 0.72 relationship with mother 0.81 0.86 0.85 0.87 0.79 0.87 0.87 connected to other adults 0.02 0.00 0.02 1.00 0.00 0.03 0.00 connected 0.75 1.00 0.81 1.00 0.00 1.00 0.85 trust in government 0.27 0.34 0.29 0.33 0.28 0.29 0.39 trust in community 0.72 0.81 0.73 0.75 0.72 0.74 0.76 spirituality 0.19 0.32 0.28 0.39 0.21 0.27 0.35 Risk Factors Average 0.36 0.25 0.38 0.27 0.28 0.25 0.25 poor family cohesion 0.42 0.25 0.31 0.31 0.41 0.31 0.29 abuse in the home 0.05 0.00 0.02 0.00 0.09 0.08 0.01 substance abuse in the home 0.42 0.00 0.04 0.04 0.15 0.03 0.08 social exclusion 0.09 0.00 1.00 0.13 0.01 0.01 0.00 community violence 0.18 0.18 0.23 0.19 0.20 0.20 0.18 felt discriminated against 0.55 0.47 0.56 0.56 0.55 0.54 0.59 live in a rural area 0.22 0.16 0.17 0.09 0.09 0.10 0.11 indigenous 0.14 0.10 0.17 0.12 0.09 0.09 0.16 low economic class (nse) 0.83 0.75 0.74 0.68 0.69 0.65 0.59 low parental education 0.71 0.62 0.58 0.59 0.55 0.53 0.47 n 237 111 169 117 137 576 134 Risk Level III III III II I - - Percent of sample 16.0% 7.5% 11.4% 7.9% 9.3% 38.9% 9.0% Notes: The highlighted rows indicate that the variables were not included in the clustering exercise. They are exogenous veriables. 36 Table 4j: Cluster analysis results: Chilean Females Aged 18 to 24 Cluster name early drop resilient spiritual connected advantaged out students mothers age 21.74 20.80 20.83 20.86 20.49 Behaviors/Consequences 0.31 0.44 0.45 0.48 0.56 Average older age when first started working 0.54 0.57 0.60 0.60 0.71 not idle 0.12 0.55 0.36 0.63 0.93 no early school dropout 0.00 0.73 0.92 0.88 1.00 older age at onset of sexual activity 0.38 0.52 0.58 0.58 0.69 older age at first parenthood 0.50 0.75 0.78 0.80 0.93 married 0.30 0.05 0.18 0.09 0.01 participate in activities 0.02 0.04 0.04 0.07 0.07 registered to vote 0.05 0.05 0.05 0.09 0.16 not victim 0.94 0.91 0.93 0.93 0.93 working 0.12 0.24 0.24 0.33 0.29 in school 0.00 0.38 0.16 0.36 0.75 years of education completed 0.36 0.56 0.60 0.61 0.68 safe sex 0.12 0.30 0.34 0.39 0.53 has a child 0.88 0.46 0.45 0.39 0.15 Protective Factors Average 0.58 0.59 0.61 0.72 0.64 relationship with father 0.57 0.53 0.64 0.61 0.72 relationship with mother 0.72 0.73 0.84 0.82 0.86 connected to other adults 0.02 0.13 0.00 1.00 0.01 connected 0.86 0.89 0.79 1.00 1.00 trust in government 0.33 0.24 0.27 0.30 0.30 trust in community 0.75 0.73 0.75 0.74 0.76 spirituality 0.30 0.32 0.37 0.34 0.33 Risk Factors Average 0.36 0.48 0.31 0.30 0.25 poor family cohesion 0.32 0.59 0.31 0.30 0.30 abuse in the home 0.01 0.56 0.02 0.01 0.00 substance abuse in the home 0.00 0.67 0.01 0.01 0.00 social exclusion 0.58 0.33 0.46 0.34 0.02 community violence 0.32 0.29 0.27 0.29 0.29 felt discriminated against 0.52 0.70 0.55 0.58 0.51 live in a rural area 0.23 0.16 0.15 0.11 0.08 indigenous 0.13 0.10 0.08 0.09 0.10 low economic class (nse) 0.80 0.75 0.71 0.70 0.65 low parental education 0.66 0.62 0.58 0.56 0.52 n 209 210 582 160 622 Risk Level III II II II - Percent of sample 11.7% 11.8% 32.6% 9.0% 34.9% Notes: The highlighted rows indicate that the variables were not included in the clustering exercise. They are exogenous veriables. 37 Table 5 ­ Ordered logistic regression for Risk Level High Risk High Risk Mexico Level Level Low parent education (1) 0.48 0.485 (0.012)** (0.012)** Being rural (rural =1, urban=0) 0.401 0.303 (0.023)** (0.023)** Age 0.135 (0.003)** Gender (female=2, male=1) 0.128 (0.019)** Observations 37903 37903 Standard errors in parentheses * significant at 5%; ** significant at 1% Figures Figure 1a. Cluster Group Means ­ Mexico Mexico Clusters Type III Type II Type I Type 0 0.80 nsae 0.60 M 0.40 ouprG 0.20 0.00 Behaviors Protective Risk Factors Rural (binary) Low parental Average Factors Average Average Factors education Figure 1b. Cluster Group Means ­ Chile Chile Clusters III II I 0 0.80 nsae 0.60 M 0.40 ouprG 0.20 0.00 Behaviors / Protective Risk Factors live in a rural indigenous low economic low parental Consequences Factors Average area class (nse) education Average Average Factors Note: The values of the group means are on a 0 to 1 scale as described in the variable construction section. Thus, the relative scale of the means between the different x values (primarily factors) is not interpretable. What is important is the relationship between the graphs for each risk level. As seen above, the Type I and 0 behaviors have a higher mean than the Type II and the Type II behaviors have a higher mean than the Type III. 38 Figure 2a. Protective Factors and Positive Behaviors Figure 2b. Protective Factors and Positive Behaviors across across Clusters of Youth Aged 18 to 24 ­ Mexico Clusters of Youth Aged 18 to 24 ­ Chile Mexico - 18 to 24 year olds Chile - 18 to 24 year olds y = 2.3288x- 0.6923 1.00 y = 0.3087x + 0.2621 R2 = 0.6628 1.00 R2 = 0.0819 sro t = 4.4 0.80 via sro 0.80 t = 0.94 ehB 0.60 via 0.60 ehB veit 0.40 sioP veit 0.40 0.20 sioP 0.20 0.00 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.00 0.40 0.50 0.60 0.70 0.80 0.90 Protective Factors Protective Factors Figure 3a. Risk Factors and Positive Behaviors across Figure 3b. Risk Factors and Positive Behaviors across Clusters of Youth Aged 18 to 24 ­ Mexico Clusters of Youth Aged 18 to 24 ­ Chile Mexico - 18 to 24 year olds Chile - 18 to 24 year olds y = -0.6138x + 0.9237 y = -0.487x + 0.6085 1.00 R2 = 0.3027 1.00 R2 = 0.2003 t = -2.08 t = -1.58 sro 0.80 sroi 0.80 avi 0.60 av 0.60 Beh Beh veitisoP 0.40 veit 0.40 0.20 sioP0.20 0.00 0.00 0.20 0.30 0.40 0.50 0.60 0.70 0.8 0.10 0.20 0.30 0.40 0.50 0.60 Risk Factors Risk Factors 39 Figure 4a. Poverty and Positive Behaviors across Youth Figure 4b. Poverty and Positive Behaviors across Youth Aged 18 to 24 ­ Mexico Aged 18 to 24 ­ Chile Mexico - 18 to 24 year olds Chile - 18 to 24 year olds y= -0.819x+ 1.158 1.00 y = -1.06x + 1.0725 R2 = 0.5993 1.00 R2 = 0.7987 sro s 0.80 t = -3.87 0.80 t = -6.298 viaheB vior 0.60 ha 0.60 Be veit 0.40 0.40 sioP0.20 tiveisoP 0.20 0.00 0.00 0.40 0.50 0.60 0.70 0.80 0.90 0.40 0.50 0.60 0.70 0.80 Poverty as a Risk Factor (parents with post graduate Poverty as a Risk Factor (parents with post education = 0, no education = 1) graduate education = 0, no education = 1) Figure 5a. Rural and Positive Behaviors across Youth Aged Figure 5b. Rural and Positive Behaviors across Youth 18 to 24 ­ Chile Aged 18 to 24 ­ Chile Mexico - 18 to 24 year olds Chile - 18 to 24 year olds y = -0.6895x + 0.8612 y = -1.3162x + 0.6401 1.00 R2 = 0.7425 1.00 R2 = 0.7465 sro 0.80 t = -6.11 sro 0.80 t = -5.426 avi via 0.60 0.60 Beh Beh veitisoP0.40 veit 0.40 0.20 sioP 0.20 0.00 0.00 0.00 0.20 0.40 0.60 0.80 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Rural as a Risk Factor (rural=1, 0.5 = mid sized town, Rural as a Risk Factor (rural=1, urban=0) urban=0) Figure 6. Chile Indigenous and Positive Behaviors across Youth aged 18 to 24 Chile - 18 to 24 year olds y = 0.1454x + 0.4203 1.00 R2 = 0.1746 sro 0.80 t = -0.186 aviheB 0.60 veti 0.40 sioP 0.20 0.00 0.00 0.05 0.10 0.15 0.20 0.25 Indigenous as a Risk Factor (indigenous = 1, non- indigenous=0) 40 References Attanasio, Orazio, Erich Battistin, Emla Fitzsimons, Alice Mesnard and Marcos Vera- Hernández, 2005, "How effective are conditional cash transfers? Evidence from Colombia" Institute for Fiscal Studies Briefing Note No. 54. Becker, Gary and Kevin J. Murphy, 1988, "Theory of Rational Addition", Journal of Political Economy, 96, 675-700. Bernat, Debra and Michael Resnick, 2006, "Healthy Youth Development: Science and Strategies", Journal of Public Health Management Practice, November (Supplement), S10- S16. Blakely, Tony, Simon Hales, Charlotte Kieft, Nick Wilson, & Alistair Woodward, 2005, "The Global distribution of risk factors by poverty level", Bulletin of the World Health Organization, February, 83 (2). Blum, Robert, 1998, "Healthy Youth Development as a Model for Youth Health Promotion", Journal of Adolescent Health, 22, 368-375. _____, R. (2002) "Adolescent Health in the Caribbean." Unpublished paper, World Bank, Washington DC. _____, R. and Marjorie Ireland, 2004, "Reducing Risk, Increasing Protective Factors: Findings from the Caribbean Youth Health Survey." Journal of Adolescent Health 35, 493-500. _____, Marjorie Ireland and Sally-Ann Ohene, 2005, "The Clustering of Risk Behaviors Among Caribbean Youth", Maternal and Child Health Journal, 9(1), 91 - 100. Bronfenbrenner, U., 1979, The ecology of human development. (Harvard University Press: Cambridge, MA). Brook, Judith S., David W. Brook, Mario de la Rosa, Martin Whiteman, Erica Johnson, and Ivan Montoya, 2001, "Adolescent Illegal Drug Use: the Impact of Personality, Family, and Environmental Factors" Journal of Behavioral Medicine, 24( 2), 183-203. Brook, David W., Judith S. Brook, Zohn Rosen, and Ivan Montoya, 2002a, "Correlates of Marijuana Use in Colombian Adolescents: A Focus on the Impact of the Ecological/Cultural Domain," Journal of Adolescent Health, 31, 286-298. Brook, David W., Judith S. Brook, and Tine Pahl (2002b) "The Longitudinal Relationship Between Drug Use and Risky Sexual Behaviors Among Colombian Adolescents" Archives of Pediatric and Adolescent Medicine, 156(11), 1101-1107. Cunningham, Wendy and William Maloney, 2001, "Heterogeneity among Mexico's Microenterprises: An Application of Factor and Cluster Analysis", Economic Development and Cultural Change, 131-156. 41 Duda, R. and P. Hart, 1973, Pattern Classification and Scene Analysis. John Wiley & Sons, New York, NY. Everitt, Brian, Sabine Landau and Morven Leese, 2001, Cluster Analysis, 4rd Edition, London: E. Arnold ; New York : Oxford University Press. Gruber, Jonathan, Editor, 2001, Risky Behavior Among Youths: An Economic Analysis, University of Chicago Press. Hall, Gillette and Harry Patrinos, 2006, Indigenous Peoples, Poverty and Human Development in Latin America, New York: Palgrave Macmillan. Hutz, Claudio and Debora Rizzo M. Silva (2003). "Developmental Issues and Delinquent Behavior of Brazilian Adolescents," International Society for the Study of Behavioural Development, No. 2, Serial No. 42, pp 6-8. Instituto Cidadania, 2004, Projecto Juventude: Documento de Conclusão (Brasilia, Brazil) Lloyd, Cynthia B. editor, 2005, Growing up global: the changing transitions to adulthood in developing countries, Panel on Transitions to Adulthood in Developing Countries ; Committee on Population [and] Board on Children, Youth, and Families, Division of Behavioral and Social Sciences and Education. Published: Washington, D.C.: National Academies Press. Milligan, G.W. and M.C. Cooper. 1985. "An examination of procedures for determining the number of clusters in a data set." Psychometrika. 50(2), 159-179. Resnick, Michael D., Marjorie Ireland, and Iris Borowsky, 2004, "Youth Violence Perpetration: What Protects? What Predicts? Findings from the National Longitudinal Study of Adolescent Health", Journal of Adolescent Health, 35, 424.e1­ 424.e10. Scal, Peter, Marjorie Ireland, and Iris Borowsky, 2003, "Smoking Among American Adolescents: A Risk and Protective Factor Analysis" Journal of Community Health, 28(2). World Bank, 2003, Caribbean Youth Development: Issues and Policy Directions. Washington, DC. World Bank. ______, 2006, World Development Report 2007: Development and the Next Generation, Washington, DC. World Bank. ______, 2007, Youth At-Risk in Brazil, Report No. 32310-BR, Washington, DC. World Bank. Zweig, Janine M., Stacey D. Phillips, and Laura Duberstein Lindberg, 2002, "Predicting Adolescent Profiles of Risk: Looking Beyond Demographics", Journal of Adolescent Health; 31, 343­353. 42 Annexes Annex A ­ Mean values of variables Mexico Chile Variable All 12 to 15 to 18 to All 15 to 18 to ages 14 17 24 ages 17 24 age 17.19 13.0 16.0 20.8 18.9 15.95 20.8 gender 0.54 0.51 0.53 0.56 0.53 0.51 0.55 Behaviors/Outcomes Average 0.62 0.56 0.64 0.61 0.51 0.59 0.47 older age when started working 0.54 0.70 0.49 0.46 0.65 0.76 0.58 not idle 0.80 0.91 0.82 0.70 0.73 0.93 0.64 no early school dropout 0.70 0.90 0.67 0.58 0.82 0.92 0.82 literacy 0.98 0.99 0.99 0.98 - - - older age at onset of sexual activity 0.78 - 0.90 0.71 0.64 0.80 0.54 older age at first pregnancy/child 0.89 - 0.97 0.84 0.90 0.97 0.86 married 0.12 0.00 0.04 0.25 0.05 0.00 0.08 participate in activities 0.21 0.22 0.23 0.19 0.10 0.13 0.08 registered to vote 0.78 - - 0.78 0.28 0.56 0.11 not victim - - - - 0.92 0.92 0.92 working 0.42 0.17 0.35 0.53 0.21 0.05 0.32 in school 0.45 0.86 0.62 0.27 0.60 0.91 0.41 years of education completeda 0.35 0.27 0.36 0.41 0.54 0.45 0.59 safe sex 0.61 - 0.90 0.59 0.52 0.78 0.35 has a child 0.19 - 0.03 0.29 0.19 0.04 0.28 low number of sexual partners in past year 0.94 - 0.98 0.92 - - - attitude towards drugs 0.97 - 0.97 0.97 - - - attitude towards alcohol 0.89 - 0.91 0.88 - - - Protective Factors Average 0.57 0.58 0.58 0.55 0.63 0.64 0.62 connected 0.94 0.96 0.95 0.91 0.88 0.89 0.87 live with both parents 0.77 0.90 0.85 0.64 - - - relationship with father 0.41 0.40 0.41 0.41 0.68 0.72 0.66 relationship with mother 0.50 0.48 0.50 0.51 0.85 0.88 0.83 connected to other adults - - - - 0.10 0.10 0.10 trust in governmental institutions 0.35 - 0.36 0.35 0.31 0.33 0.29 trust in community institutions 0.61 - 0.59 0.62 0.76 0.79 0.75 church attendance 0.38 0.42 0.38 0.35 - - - school quality 0.80 0.81 0.80 0.79 - - - optimism about future - - - - 0.88 0.88 0.87 preparation for future - - - - 0.84 0.79 0.87 communication with parents 0.48 0.51 0.46 0.46 - - - spiritual influence / spirituality 0.19 0.17 0.19 0.20 0.33 0.36 0.31 sense of wellbeing 0.85 - 0.84 0.85 - - - Risk Factors Average 0.56 0.57 0.56 0.56 0.28 0.25 0.30 parental influence (alcohol & smoking) 0.49 0.37 0.44 0.60 - - - poor family cohesion - - - - 0.34 0.34 0.33 abuse in the home - - - - 0.06 0.06 0.06 substance abuse in the home - - - - 0.09 0.07 0.10 social exclusion 0.53 0.67 0.54 0.43 0.18 0.10 0.23 limited access to healthcare 0.49 0.50 0.51 0.48 - - - community violence - - - - 0.25 0.26 0.25 parental response to misbehavior 0.28 0.29 0.26 0.29 - - - parental response to good behavior 0.55 0.52 0.54 0.57 - - - felt discriminated against - - - - 0.55 0.54 0.55 43 indigenous - - - - 0.11 0.11 0.10 live in a rural area 0.43 0.46 0.44 0.39 0.13 0.13 0.13 percentage living in rural area 0.25 0.28 0.26 0.22 0.13 0.13 0.13 low parental educationa 0.69 0.69 0.69 0.69 0.57 0.58 0.57 household ownership of goods 0.78 0.79 0.78 0.77 - - - monthly earnings household heads 0.80 0.80 0.80 0.81 - - - low economic class (nse) - - - - 0.54 0.29 0.70 Sample size 37979 11871 9438 16670 5321 2057 3264 * highlighted rows are the variables that were treated as exogenous (they were not actively clustered on). Notes: aIn Mexico, a value of 0 indicates no education, 0.17 would be exactly 6 years of education, 0.34 would be exactly 9 years of education, 0.51 would be 12 years, 0.68 would be 15 years, 0.85 would be 19 or 20 years, and 1 would indicate a post graduate education. In Chile, a value of 0 indicates no education, 0.15 would represent from 1 to 7 years of education, 0.3 indicates a primary graduate (8 years completed), 0.45 from 9-11 years, 0.6 indicates a secondary graduate (12 years completed), 0.75 would represent from 13 to 16 years, 0.9 indicates a tertiary graduate, and 1 indicates postgraduate education. 44 45 me ho ine cen home de abus in ralur) ec violey ed clud home ncea home use nit ation sione abuse violenyt (20%) exy in in ce most e e us us use bstus us erty us ugs r use nott abe mmuoc udclxe coh unim mination noe parents Factors (27%) sociall ab ab substan (bu digin most hom stom ralruts digenousin pov abe dre riminc digenousin poo ralru least um dis hom hom st most least socially family com household discri least st thiest Risk Rural Indigeno Feel Most Poorest nd 2 Most Most More No nd 2 Lea nd 2 Medi No No Lea nd 2 nd 2 0% Best No Most Low nd 2 Lea Weal low as erthfa er & father not withp nte future ty moth ection (but withpi employmente erutuftu ghlyhiyt erutuftu retuuf conn hersto abo ict for vernm dte lot)a aboutc go (by communi withtc alt Factors ected )rs abo unim to edar ec t relationshi timis spiritual int conn istim retufurofder ualt int paren epa men connected ective conn loneeth relationsh positiv com op st st reppts usrt opti connet least feel prt spiri usrt connected erthaf most Prot Low as Poor Poorest Most outlook Optimistic Tru connected Lea Lea nd 2 Low 0% Most Bes Most Most govern All Highes and Highly nd 2 ) g xse rking activity r & & group in out in on rking ate sexual ped sexe oldetr tiesi voteot lient" working ate dna wo of re "resi te educati artst particip dropt wodteratse ag es onsety no xse unsaf um most Behaviors Lowest Young Lowest activiti Earl Have Medi Young nd 2 stadnaxserefaS workintratse ag sex start to particip vo vote um sextrats tivca desit safe lowest agere es sex workingdtearstt Most Medi and nd 2 Lowes (similar 87% Old sex Higher activiti stefaS um ipationc deslO Medi parti ) group (3%) (36%)g (98%) (4%) school child ol lient" s child scho "resi esm ledi workint high with in dski with cluster ledi to school school school working kid working ledi most least ledi working kids in working have in working the 15-17s Outco Most Dropout Highes Most In Not 8% No 5% Not Finished nd 2 nd 2 Not 8% (similar No All 2% 1% All 1% of Male pu pu gro pu gro doing gro group group group doing so & Description­ Chilean age-sex B B1: IIIe age-sexeth IIe the ected of Ie age-sexeth conn Ie age-sexeth of 0e connect age-sexeth of 0e age-sexeth typ of typ typ of of typ 105) typ typ =126) Annex Table Cluster At-risk Risk 7.4% (n=75) Risk Resilient 35.3% (n=359) Risk Other OK 8.6% (n=88) Risk Loners 10.3% (n= Risk Parental well 12.4% (n Risk Advantaged 26.0% (n=265) 46 e d in ec abuse us us use househol violenyt tiona (20%) clusion stanc ) ex violentt ion sione ions abuse clusionxe cial Factors 5%)2(lraru abec exclusion in tan 18%)(la us clusionxe cohesion meoh in (8% cial mos y clus coh hh hh cial use d unim rur geno sube digenoinro digenoinro exclu ines so ab subs com discrimin indi cial family so hom rural family rural cial so st most abused ralruts so st 1 ex in most unitm social abuse so abu household Risk Most Poorest Most Most Some househol Lea Most nd 2 Most Poorer Low Worst All Most Lea Poorer nd 2 Reports com Low no best no Low No No No High erutuftu dna er or t retuuf nt fath adul ty father, ty menr ty nt parents, adultsre future adultsre future other on abo with menr with on erutuftu retuuf 41%)( father, other oth oth ict Factors forder abo to gove hip for ed withpi to aboutc to aboutc ualityt mmuni with ewiv veog to ict mmunioc therom communi in ve timis epa spiri op pr intsutr co ect ewive in tions istim ualt int istim ualt connect sitive conn po positiv op opti spiri connected opti spiri st most ts intsutr y edar timis with ection unitm rela ther connect st relationsh usrt connected mo st reppts usttrts onnectc connected connected conn connected Protecti Lea Most Lowest Most nd 2 Poorer some adul Least employment astle com all worst or most Most employment Lea Lea Lea weoL KO Poor but Not Most Most Most All Most mother No Most Most All and es &g sex sex (82%) & activiti ex ex xese rkowtr teov safenu workingt in sex workin rk/s rk/s tot sta safts ste tivitiesac wan mostd staregae ipatec safe start rt parti repo when sex wotarstt wotarstt st Behaviors Lea Earli Lea Most Secon Middl sex Most Many Older sex stefaS xes deslO Safer deslO y attainment earl out onal (93%) childa ed 15-17 m es school (93%) children kida rking kid rking kid rking with in educati dropp ledi working with school working ledi school wo ledi school wo wo school in have in with in with in age Outco Half None 99% 7% Lowest Most Few All Most None 5% All None None 2% All None 1% None All females pu pu with re with gro dte gro cially group oth group Chilean ecn nnected witht nnected (co (co B2: IIIe age-sexeth IIe cont bu age-sexeth Ie sodnadte ec age-sexeth age-sexeth 0e typ of 95) typ of 60) typ ded of 0e 133) typ notubst of 661) typ clu = ts) = Table Cluster Risk At-risk 9.1% (n= Risk At-risk 5.8% (n= Risk Disconn ex 12.8% (n Risk Advantaged paren adul 63.6% (n Risk Advantaged all) 47 me d ho in edd (9%) househol abuse in ation cohesion ce ytr uded us excl ytr cluxe us cohesion use Factors lraru riminc geno noe family abec substan dis rural povee ralur indi povee socially st socially ralruts ralruts household digin family in st Risk Most Poorest Worst Most (42%) Lea High Middl All High Most Middl None Lea Lea Lea Worst Abuse Substan (15%) er to r er or ts fo d an moth nte fath father prepd ther adul nte to on an mo prep connected with erutuftu retuuf on on with all ewiv vernmog (75%) abo with with other ict future, to ewiv least and vernmog and Factors in ewive connected y forder in y er sitive timis ualt sitive adults po spiritual connected relationship epa re po spirituality usttr connected relationship most fath ective usttr st unitm op st connected relationship pr positiv utuf spiri connected relationship connected re st unitm nd 2 and other Prot Least employment Least Lea com Least Worst Lea All Best Most Most employment, for Most All Best All Least employment futu Low Lea com None Worst mother es es es ex tivitica ex tivitica activiti xese safts work/starsttse ine rk/s (8%) ine in d te rrie ipatc mats artip sex wotarstt d rrie ipatc vo artip st Behaviors Lea Young Lea Lea stefaS married st highest sex participate vote sex deslO mats Lea Most Lea nd 2 Safe High High Safe drop level (13%) secondary on level (57%) school rking school esm wo kid kid aryd school finished (28%) educati rking school dropy 18-24 in with wo in education working with idle school idle earl most idle finished working school in in secon age Outco Few nd 2 22% Fewest school Lowest None None All All High Most 25% 20% ½ 20% 15% 1/3 ½ No pu males pu pu pu gro group gro group gro gro ate age-sexeth Chilean B3: IIIe age-sexeth se age-sexeth upbd of of IIIe age-sexeth IIIe Ie age-sexeth Ie age-sexeth 90) typ 237) typ aduatgr ultsda of of an 111) typ 169) typ of 117) typ of 137) = = = = = = 8.7% (n Table Cluster Risk At-risk 16.0% (n Risk Idle 7.5% (n Risk Young 11.4% (n Risk Active 7.9% (n Risk Resilient 9.3% (n 48 ) (58% ec home (8%) edd in clusion clusion us noe ded (16%) ded (13%) us cluxe violenyt tiona 16%)( ex me erty cohesion ho abuseec ex me erty erty ho ralruts household digin in cluxe r unim in discrimination cluxe geno cially ralru pov social ralru pov social pov indigenous poo indi so com discrimin family tans ines st poor Factors 3%)2(lraru um um um used ab sub um um um ralur um abu Lea Lea Low Abuse Socially Highest High Least Socially Risk Most Most Poorest Most Most Most Medi Medi Medi Worst ½ 2/3 Medi Medi Medi Low Medi No ther ,eru fath adultre mo fut ,erutuftu with nt menr with erutuftu er nte oth of rall)e on with thetu with the abo abo to (ov etc... abo int ict gove relationship ict vernmog t ewive ewiv spirituality usrtt Factors connected futurer rst in ected retufurofder timis fo timis er ualt sitive men relationship relationship epa connected connected um op spiritual wod conn ective intsutr op usttr connected po st prep st st moth positiv spiri prt connected All Best Optimistic employment, High Highes govern Medi Prot Lea or Least Most Worst mother Secon father Lea Lea Worst and Some (13%) Most employment Most Least Least employment Bes All &g es xse xes & and xse activiti in (30%) votet xese & ctimviagineb workintratse work ag ipatec sex workingre working iedr ste married report sex um mar sex married parti Safe Old Highes derlO safts Behaviors Lea Earli Most 6% safe Medi sex 5% safe 18% Most en dr dropout chil dropout yard (64%) ationcude with school school level e school econs (33%) school idl aryd mber kida ed working in mostd es school school kida school kida not secon working education idle nut 18-24 m working in with idle secondary education in e finished in ledi finish working 40% Most Secon Most No 40% Most 20% Lowes age working with idl working with Outco 12% None 88% 88% All Low ¼ 38% ½ ½ 73% ¼ 16% ½ 1/3 Most school Most s pu females group gro group group havioreb esm pu group gro outco 0e age-sexeth voters artt(s of 0e age-sexeth Chilean but B4: IIIe age-sexeth of IIe age-sexeth t of IIe age-sexeth life IIe age-sexeth typ 576) typ of ultsda of 134) typ 209) typ 210) typ in = = = = r)e 582) typ of 160) = Risk Advantaged 38.9% (n Risk Advantaged 9.0% (n Table Cluster Risk at-risk 11.7% (n Risk resilien 11.8% (n Risk young later earli 32.6% (n Risk Connected 9.0% (n= 49 ce (58%) to drugs to edd to influen de care ce edd cluxey me uded health ho exclusion access drug response havioreb cludxe fluenni response r to cluxe r ralruts ines alt alt bad excl ren poo r r poo sociallts abu Factors social limited care pa paren and lraru socially um cessca socially ntalearpts rentalap poo socially ralruts poo Lea Least Lea No Risk Some Most health Most Most good Most Poorest 58% Medi All Most Lea Least misbehavior Least None Lea Least ts st er ther adul future fath e ther mo paren anc mo e h ce with parents wit with with with other aboutc with both nde with ancd att iont quality influen both chr with istim Factors with ch tenat th quality quality er ch ualt chu wi icanu to opti relationship school relationship relationship connect connected relationship erthaf live st churt ective connected moth relationship erthaf chur spiri live st mmoc st go st school st school connected All Most All Best and Prot Fewe (87%) Worst and Lowes Worst Least Best and Most Most all Most (93%) Better paren steB connected paren Most paren Best All Better paren Best sex es & g rk rking wo work wo activiti workin start 53%)( %)1(d in rrie ipatec start age ipatec d start rke age sex when stefaS mats parti vote artstegatse um parti wore um Oldest Lea Most Most Behaviors Medi Young Most Nev Medi es )e dropout ) tivitica (75%) edat (15%)dkia idlera school (93%yc ine school uc th (7% aryd school ipatc rking in ed wi st ledi 12-14 esm e artip school wo school ledi school ledi secon working in idl eralitt st working ledi working in literate in literate in literate Most Most Lea Not No age Outco Half None Half Lowes Lea 42% All None All None All None All 20% All None All males ers loners) group group worky group but group group 0e age-sexeth Mexican earl but of B6: IIIe age-sexeth of IIe age-sexeth of Ie age-sexeth (advantaged, of 0e ed age-sexeth typ 622) typ typ 1193) typ 1893) typ agt of 1989) = Risk Advantaged 34.9% (n Table Cluster Risk at-risk 13.0% (n=758) Risk connected 20.5% (n= Risk loners 32.5% (n= Risk advan 34.1% (n= 50 to to drugs drugs to to de to to re ce ce edd tos to cludxe access response uded uded or fluenni response alt access response fluenni access excl r healthca response xclue excl alt cesacd po excluded alt alt behaviord r good Factors socially limited care paren havioreb lraru alt poo tos cially ren limited care paren pa paren lraru limited care socially um gs socially acces ntalearpts paren havioreb ralruts poo socially miteli care asteld Factors so an Risk 73% Most health Most good Most Poorest All Most health Most dru Medi All All Lea Most good Lea Least None More health Secon Risk Half Most Most mis- Most Poorest Most health st er st re fath ren moth dnah ) s er 91% fath nt paren with pa father wit parent still and er with with variables parents st with st menr both ication withs with with ut(b ication better fath quality hip the ionst both Factors with all th icanu arenp un arenp quality gove well-beingfo Factors with with er live mmun tions on wi er with er mm with sense relationship co ective moth st tivitieac school rela es connected es school st ertt um live relationship moth mmoc relationship co relationship ective moth Prot Fewest (14%) Worst and Worst paren Fewe Worst Be Medi Most (93%) Best and steB intsutrtsro activiti Best Prot Least Worst and Worst activiti Worst W Lowest Connected relationship g s g ug es drsd workin tivitica rking se work towar school started ine ine xesefsa school in tiviticat workintratse ag rlyea school in um school school in in um detuitattsro non 15%)(xese when ipatc wodteratse ag saf alcohol st xes artip (93%) st um Behaviors None Lowes All Medi All All Medi W and Behaviors Almost Lea Youngest & Lea Most Medi es tivitica (27%) (8%) ine level 13%)( age (70%) on 12-14 (72%) ipatc age esm 15-17 educati working ledi illiterate artip working ledi work st working ledi literate literate ste working ledi esm working idle school school literate age illiterate in working in Outco Most Most Most Lea Some None All 17% All None Old 11% All None Outco Most 25% 3% 6% Lowest 37% All females males group group group group group Mexican B7: IIIe age-sexeth uth ers loners yo Mexican of IIe age-sexeth of Ie age-sexeth of 0e ed age-sexeth ag of B7: IIIe age-sexeth IIe typ 908) typ = worky of 2723) typ 982) typ 1376) typ 1765) typ risk = Table Cluster Risk At-risk 15.2% (n Risk Earl 45.5% (n Risk Advantaged 16.4% (n= Risk Advant 23.0% (n= Table Cluster Risk At-risk 39.8% (n= Risk High 51 to th drugs to ) wi e and re ce edd re to ce ded clu car response uded althca althca access lo t exy influence, excluded he fluenni response cluxe he (68%ded clusionxe fluenni health r excl res ciall to misbehavior r to riovah cially parental tos tos cluxe ste alcoh cial ralru be so poo socially cesac ntalearpts rentalap socially cesac poo ralruts parental Factors lowd care lraru poor so parental andsg um pood soyrev ssecca Half Higher behavior More All All Lea Least behaviors None All Least Lea Risk Socially Secon health Most Most High High dru Medi Secon Not Low Worst response odogot lrarueld Mid Poor r t with thefa & ce chr men th father nt parents with ce quality with parents ch parents ofe bo with chuotg with govern and with int well-beingfo menr influen or1 ship influen es gove both relationships chur quality wellbeing both shipsn sens school Factors th of th tioa ste connected with ualt ste goint st usrtt er sense er wi wi cenadne relationship moth tivitica lationer moth intsutr ste ste rel spiritual school sense live relationships spiri att ective connected live lowd parents lowd live um st um lowd asteld st st ch lowd Highes paren Lowes Best and Most communication Highest Higher and Most Prot Well Most Secon with Secon attendance Worst Worst Connected Most Medi paren Medi Secon wellbeing Secon Fewest paren Worst paren Lowest chur Secon last ingk xse dtearst dtearts th havioreb mostd wi the inr in xestarst (90%) wortartse working xse enhwe activityfo en (15%) sex ualxese an activity mostd rtne work al sex an start (90%) start start pa1 ag ag & velel whe ag /saf tivitiesac sexu rlyea ste sex ste agere sex agere agere involvement um um ipationc ste um um ste having than um es r Lat Safer Old Old Safer Old Old Behaviors Medi working Medi parti Earli working Medi Medi Earli Most unprotected more yea Medi Lowest activiti le le ) lev lev ) level ation ation (96%yac (95%yc on drop early uc ed st uc ed liter out 15-17 st ste paren ste ste serh ledi school ed working school ledi working school ledi school ren educati schooly literate in literate in literate age esm eralitt in ledI in are school pa lowd lowd lowd idle married working earl working working dropp in motear None All 27% All None All 27% All None All Outco Idle Not Lowes All None Lowest All None All None All Secon None 17% 6% Secon Secon 80% 54% 66% females pu group group group group group gro age-sexeth Mexican ame?) of Ie age-sexeth (ren of 0e age-sexeth ers of B8: IIIe dropouts age-sexeth of IIIe age-sexeth IIIe age-sexeth typ 681) typ 869) typ early 932) typ worky of wives 603) typ of 363) 25.3% (n=1120) Risk Loners 15.4% (n= Risk Advantaged 19.6% (n= Table Cluster Risk Idle 18.7% (n= Risk Earl 12.1% (n= Risk Young 7.3% (n= 52 d reca to an drugs response and and (37%) reca rds (low (36%) to ions health ions coholal to rural uded health r response haviore towaec to )% ssecca exclu cial ssecca ralru parental coholal poo exclu ste exclusion exclusion r alt reec excl r re uen infl smokingd high rst an so um um cial lowd poo so paren havior/misbehavioreb poo d influen social wod care gs socially ralruts stebd ssecca goo Factors social (31larur No Worst Medi Medi No Secon Least Best good and dru All Lea Least Secon to behavior/misb influence Risk Some Worst Parental alcohol variable involvement) Most Poorest Some Secon health d ant y) st st er ht & y) men ingeb with with with ce llew happ (bo paren paren moth ngi parents e munit anc well govern of int institutions happy)gnieb of ingeb ofe with with llbe with tionshipsa influen quality rto both th quality tionshipsa quality we nde att rel ualt sense on comd on rep wi wellbeingfoe rel senstes relationship stitutionsni of usrtt sense rt rst in ch connection spiri att school live school school Factors ant um repo stebd cenadne st nsese ch um/low (82% nsest relationships stebd st um highd connecti wod relationship trust men ective st connecti relationship churt Lowes community Medi (85% Lowest Secon paren Highest chur Highest Medi being Highest Best Highest Highes Secon paren Medi Secon wellbeing Prot Good Secon father Worst Worst govern weoL (82%) Good Worst Lowes ) coholal drunk sdar job job tow job de first vior getting s at ha es first jobtsrif bela s rstfi towards behavior at behavior at s ttitua ate tivitieca age age justify rst ag um xuse tivitieca age sexual tiviticat um sexual attitude sex can um tivitieca wod sex st High Medi steB married um gs High Oldest Good Highes Medi Good Worst (11% Behaviors Medi Low Secon dru 14% Medi genuoY married 14% y ) y literate literate literate earl earl dropout fully fully fully level out out s,t early s,t s,t ed (90%yc ed ledi school et working in paren ledi school working in 4% paren ledi 18-24 working paren esm age idle education dropp litera eralitt dropp ledi Not 23% 95% Not Not 14% 95% Only Not Not 17% Not Outco All Low 73% 99% Lowes 50% Not males group group group group group rse Ie lon age-sexeth uth loners yo Mexican of Ie age-sexeth of 0e ed age-sexeth ag of B9: IIIe age-sexeth age-sexeth typ 957) typ 947) typ 1178) of IIIe typ 743) of typ Risk Resilient 19.2% (n= Risk Advantaged 19.0% (n= Risk Advant 23.7% (n= Table = Cluster Risk Idle 10.1% (n Risk Drinkers 20.6% 53 to e to ce ce ce (27%) sion car alt d alt fluenin ralru t cluxe altheh tos influen response alt alt secon lanter res al tos ntearp r parentes sd ions influen response oorestp gooy r cesca exclu ren behaviordab ded parentes far)y(br parental mostd pood socide poo cescat paren parental to cluxe um behavio rural highd xurlu cial /thirdd pat havior,eb highd so um stebd care Low Secon Secon Limit (10%) Bes Medi Worst poor 20% Secon education Low No Secon Highes Highest good highest Socially Secon influence Medi Secon health pateshgihdrihT poots influence Lea t snte d st par withe ce father an ofe paren with anc ce nte paren ty with nde both with both quality sens on neither chr ste th chu wellbeingfoe with shipsn influen quality wellbeing att of connectedness or1 fluenni or1 mmuni ch ualt institutions th quality vernmog th ndanceetta quality co wi wi er y wi school lowd to connecti live go nsest tioa rel school in ofle sense tionshipsa int llbeingew chur int um st churt st spiri lev um trust um live st rel moth spiritual school usrt unitm um live st relationships ch school usrt Worst Secon wellbeing Good Most Most Highes Connected Medi paren Lowes paren Lowest Medi Low Medi Lowest Most paren Good and High High High com Medi Connected Most paren Best High High High sdar s activities tow ied)r xse ug partners work sexual in de sexual mar workt ofte drsd job xese of &sgrud on tose ttitua sex of %(5 sexual of towar activities activities rstfi startta rst saf iedr tivitiesac startae ag onstae ag mber ofle at married un number ofle rse participati tuditat wod rliestead number rse xsetsfesa age sex gs mar um um lev lev destold 99% Most Worst partn Low steB worky alcohol Secon dru Earl Secon Worst partn Not Medi ndoceS ste nut detuitattsro Medi Old Bes Low W (77%) High Oldest Secon stefaS ed hievca ut le level level lev on velle d dropo level chila school educati school dropy education children education children working in school schooly working in education have earl high earl working education ledi ledi school working in tera um have in working ledi have 80% 30% Lowest All 7% Low 71% All 91% Low Not Not 78% 28% 75% Lite Medi 11% Very 74% 56% None 3% group group group group ts ds IIIe da age-sexeth rse endust of IIe age-sexeth of Ie lon age-sexeth of 0e ed age-sexeth 1511) typ 837) typ dropoutsy ag of 1778) typ 1335) typ 1142 (n= Risk Working 11.4% (n= Risk Earl 24.2% (n= Risk Resilient 18.2% (n= Risk Advant 15.5% n= 54 reca de altheh reca e care car smoking) from cludxe tos over 26%)( parental or poor andl cescae r ralru ofle ded healthot control clusionxe health altheh to poo tos po lev cial ralru ndente r Factors cially ted cluxe so hav um um cesscat 6%)3(lraru so access ralur cescat asteld um um/high indep (alcohost uralrt poo Risk Not Half Medi Medi Highest attemp behaviors Socially Lowes Most Poorest High Poor Low Bes Secon Medi Medi Most paren Lowes Least st nte ren ce pa ce steB ce both of ual with parents" with ntse with ntse governm orst in oreno with parents andne parents relationships influen quality veil th shipsn sse shipsn att th influen wellbeing Fac rest ualt leveltes ed with /spirithcr parht parht wi ch wi wi ualt usttr eing wie tioa tioa of with spiri onnectc "live chu relationships school connected /high liv rel happin livet rel churt livet spiri um wellb live st pood parents highd st um st um happiness um um um um st um st sense Medi High Protective Few paren Connected Secon with Lowest Secon happiness weoL Medi Worst paren Medi influence Lowest Low Medi Medi Connected Medi Medi paren Highest Connected Highes Medi paren Highes Highes relationships Highest High yc sdar use ug xes arey egnanrp tow coholal coholal de drsd sexe rstiftae past in first ate se ttitua se towards work towards towar rse sex xeseaf ag rtn work /uns saf ste pa agre tiviticat rst wod safe exsy tiviticat attitude rliestead sex tivitiesac um um ste um sex attitude sex Behaviors Low Earli Most Young Lowes Secon alcohol Average Earl Lowes Worst Secon Medi Medi Earli Medi stefaS Worst stefaS detuitattsro W in t yr 2% when this (ve school school Lowes in school (90%) in in out (evenle s)le level level level level level out on ut out on 18-24 dropy lev maot rking,ow %92 %84 %96 ati children (5% illiterates illiterate) child educati all dropy child uc edt age earl idle Outcomes All 93% education edarpmoc dropy (8% dropoy education education education have married idle earl married earl ledi working, with um arle with um married ledi working, um ledi working, 68% 63% 93% school) 81% Lowest Almost group 40% All Not 83% 23% Medi High 40% Medi 45% Not 60% Medi Not 40% Highes high) males pu group group group ok group gro group ts Mexican uts endust B10: IIIe age-sexeth doing r of IIIe age-sexeth loners of IIe dropo age-sexeth of IIe ers age-sexeth of Ie poo age-sexeth 0e ed age-sexeth typ 1810) typ 2324) typ typ = worky 999) typ of ag of 850) typ 1644) Table Cluster Risk At-risk 19.4% (n Risk At-risk 24.9% (n= Risk Working 18.2% (n=1693) Risk Earl 10.7% (n= Risk Resilient 9.1% (n= Risk Advant 17.6% (N= Annex C ­ Kernel density graphs of parental education level by risk level The figures38 below demonstrate that the more at-risk clusters in both Chile and Mexico have a lower level of parental education than the other clusters. The education level of parents or heads of household = 0 if no education or less than a primary school education, 1 if primary graduate, 2 if secondary graduate, 3 if tertiary graduate and 4 if above tertiary level. In Chile, parental education level is more strongly related than the assigned socio-economic level (nse)39. This could be due to unobserved characteristics related to educational achievement. We also see that Chile has a higher parental education level than Mexico. Chile Kernel Density diagram of parental education level by risk level Parental Education Distribution for Each Risk Level 5 1. 1 yits Den .5 0 None or less than primary Primary Grad Secondary Grad Tertiary and above Education Level of Heads of Household Risk Type III Risk Type II Risk Type I No Risk Mexico Kernel Density diagram of parental education level by risk level Parental Education Distribution for Each Risk Level 2 5 1. yit ns 1 De .5 0 None or less than primaryPrimary Grad Secondary GradTertiary and above Education Level of Heads of Household Risk Type III Risk Type II Risk Type I No Risk 38Generated using kernel density plots with Epanechnikov kernel. 39Details available upon request 55