Adaptation of the Communities That Care Youth Survey for use in Brazil: A pilot study

Summary

Substance use and violence are serious problems affecting Brazilian youth, prompting local authorities to invest in adapting and implementing evidence-based prevention programs developed abroad. To ensure that interventions are effective, valid and reliable measures of risk factors for substance use and violence are needed. We conducted a pilot examination of the measurement properties of risk factors in the Communities That Care Youth Survey (CTCYS), a school-based survey developed in the United States. The Brazilian-adapted CTCYS included 25 risk factors as well as measures of substance use and antisocial behaviors. The survey was translated into Brazilian Portuguese and adapted for use in Brazil, incorporating suggestions from local experts. The survey was administered to 926 students in one Brazilian community. Results indicate that scale internal consistency was acceptable for all but two scales: (i) Academic Failure and (ii) Rebelliousness. Confirmatory factor analyses suggested good model fit for most measurement models albeit with minor modifications for most scales. Findings from this pilot study provide a foundation for the use of the CTCYS in Brazil to assess community levels of adolescent risk and evaluate prevention programs.

INTRODUCTION

Youth drug use in Brazil is alarmingly high. In a study with over 50 000 Brazilian students 10–19 years of age, participant reports of past-year drug use were 42.4% for alcohol, 9.6% for tobacco, and 10.6% for illegal drugs (Carlini et al., 2010). Rates of youth cocaine and inhalant use are some of the highest in South America and are similar to those found in the US (Carlini et al., 2010; Johnston et al., 2017). Of great concern is the early onset of drug use, with youth having had their initial contact with alcohol and tobacco at age 13, and first contact with other drugs between ages 13 and 15 (Carlini et al., 2010). Youth violence also represents an urgent matter for Brazilian authorities. In 2015, over 23 000 Brazilian adolescents were incarcerated or had their freedom restricted (Conselho Nacional do Ministério Público, 2015). Juvenile justice data from 2016 indicated that 46.62% of youth had been incarcerated or had restricted freedom for theft charges, 22.50% for drug trafficking, 9.82% for homicide, and 21.06% having been charged with other serious crimes such as rape, possession of a firearm or attempted homicide (Ministério dos Direitos Humanos, Secretaria Nacional dos Direitos da Criança e do Adolescente, 2018).

In light of both national and international evidence, Brazil has been increasingly interested in prevention, with local authorities investing in preventive interventions in the early 2010s (Murta et al., 2021a, b; Sanchez et al., 2017; Schneider et al., 2016). Although measuring impacts of preventive interventions on behavioral outcomes is important, it is crucial to understand how interventions affect the proximal causes of targeted outcomes. Additionally, prevention program development, planning, and evaluation should be based on risk and protective factors (RPFs) that have been identified using rigorous longitudinal research.

The Communities That Care Youth Survey (CTCYS) is a self-report questionnaire that collects demographic information and assesses youth exposure to RPFs and drug use, delinquency and violent behavior outcomes (Arthur et al., 2002; Brown et al., 2009). The CTCYS assesses RPFs in four different domains of youth life: community, school, family and peer-individual. The CTCYS is administered typically as part of the of the Communities That Care (CTC) prevention system (Hawkins et al., 2008). Specifically, it serves as a planning tool to help local leaders develop a community profile, identifying what changes are necessary to ensure youth bonding with the community and healthy youth development. It is also used to monitor changes in levels of community RPFs resulting from program implementation. The CTCYS has been adapted for use in different countries; for example: Australia (Toumbourou, 2017), Chile (Varela et al., 2020), Colombia (Pérez-Gómez et al., 2016), Germany (Groeger-Roth, 2017), Iran (Baheiraei et al., 2016), Malaysia (Razali and Kliewer, 2015), the Netherlands (Jonkman et al., 2005), Pakistan (Khalid and Kausar, 2016), South Africa (Morojele et al., 2002), and Trinidad and Tobago (Maguire, 2013).

The objective of the present study was to conduct an exploratory psychometric study to ascertain whether risk factors for youth drug use and violence can be measured in Brazil using established measurement models developed by researchers in the US (an analysis of protective factors is beyond the scope of this article). In a broader sense, we aimed to develop a prototypical instrument to assess levels of youth risk in Brazilian communities, thereby providing local community stakeholders with a better assessment of needs and priorities while also contributing to the advancement of prevention science research in Brazil. To the degree that RPF measures have comparable construct validity among different countries, these measures can increase confidence in the cross-national use of preventive interventions that were developed to address these RPFs.

METHODS

Participants

Participants in this study were 926 Brazilian students from two public high schools in the community of São Sebastião-DF. The inclusion criterion for participation in the study was being enrolled in one of the two participating schools, under the regular curriculum (i.e. not special needs, accelerated learning or adult education curricula). Official school enrollment rates were 1683 students for School A and 1649 students for School B. Because of elevated school dropout rates only 946 and 955 students, respectively, were present in the schools during study recruitment. Response rates were 56.45% for School A and 59.37% for School B, which was due to students refusing to participate, forgetting to obtain parental consent or being absent from school during data collection. Some participants were removed from analysis due to failing validity checks or not complying with instructions (e.g. sharing answers with others). Additionally, 176 participants had to be removed from analyses because of questionable comprehension of items. The final sample consisted of 908 students enrolled in ninth, 10th and 11th grades (ns = 289, 344 and 275, respectively). Participants’ median age was 16 years, with 3.20% from 12 to 14 years, 76.27% from 15 to 17 years and 20.52% 18 years or over. The majority of participants were female (58.82%) and resided in urban areas (86.15%).

Measures

The Brazilian-adapted CTCYS included questions on students’ sex, age, number of siblings, parent education and where they lived. The instrument measured 25 risk factors for youth drug use, violence and delinquency, assessed through 108 specific items. The survey also measured protective factors, but for the purpose of this article we will only focus on risk factors. Risk factors included in the adapted survey were based on the 2012 version of the US CTCYS [Arthur et al., 2002; Social Development Research Group (SDRG), 2012].

In the community domain, Low Neighborhood Attachment reflected the extent to which youth felt they belonged in their neighborhoods. Community Disorganization addressed neighborhood characteristics such as safety, physical deterioration, public displays of violence, and high rates of adult crime. Transitions and Mobility asked whether participants changed schools or homes frequently. Perceived Availability of Drugs assessed participants’ perceptions of ease of obtaining cigarettes, alcohol, marijuana, and other illegal drugs in their community. Perceived Availability of Guns asked about participants’ perception of ease of obtaining firearms in the community. Laws and Norms Favorable to Drug Use focused both on community residents’ attitudes and norms about drug use, as well as law enforcement action in response to youth drug use and delinquency.

In the school domain, Academic Failure addressed failure to achieve passing grades. Low Commitment to School assessed aspects such as liking or hating school, perceiving schoolwork as interesting and relevant, making an effort to do well at school, and skipping class.

In the family domain, Family History of Antisocial Behavior identified whether the participant’s family had a history of criminal activity, alcoholism, and cigarette or other drug use. Poor Family Management addressed inefficient family management practices, such as unclear expectations for behavior, poor parental monitoring and failure to check youth’s homework. Family Conflict measured the level of conflict among family members, namely yelling and arguing. Parental Attitudes Favorable Towards Drug Use identified families in which parents tolerated youth drug use. Similarly, Parental Attitudes Favorable Towards Antisocial Behavior assessed parental tolerance of youth delinquent or violent behavior.

In the peer-individual domain, Rebelliousness focused on whether the youth questioned authority, ignored rules, or actively opposed them. Early Initiation of Drug Use addressed early onset of cigarette, alcohol and marijuana use. Early Initiation of Antisocial Behavior covered early onset of violent and delinquent behavior. Gang Involvement concerned youth’s gang membership and how many of their friends were gang members. Favorable Attitudes Towards Drug Use tapped into participants’ values and attitudes that favored drug use. Similarly, Favorable Attitudes Towards Antisocial Behavior pertained to participants’ values and attitudes that favored engagement in delinquent and violent behavior. Perceived Risk of Drug Use assessed participants’ perception of negative consequences of both occasional and regular drug use. Friends’ Use of Drugs referred to youth being friends with other youth who used drugs, alcohol and cigarettes. Interaction with Antisocial Peers addressed youths’ friendships with peers who engaged in violent behavior or delinquent activities. Rewards for Antisocial Involvement asked whether youth believed they would be admired by their friends for using drugs or carrying firearms. Intentions to Use Drugs assessed participants’ intentions of using drugs in adulthood. Additionally, a Sensation Seeking scale measuring engagement in risky and thrilling behaviors was included in the Brazilian adaptation of the CTCYS. The survey also included 24 items that assessed drug use frequency and 22 items assessing antisocial behaviors, such as suspension from school, selling illegal drugs, stealing and being arrested.

Procedure

Translation

Adaptation of the CTCYS for use in Brazil followed the steps proposed by Borsa et al. (Borsa et al., 2012). Items were translated by two independent bilingual translators, translations were compared, and decisions were based on best correspondence and parsimony. Translated items were presented to ten field experts (community leaders, residents and professionals working with youth) who provided input on items. Additional items pertinent to youth problem behavior in Brazil were incorporated into the survey per expert recommendations (i.e. domestic violence, economic deprivation and receiving threats). The adjusted survey was reviewed by 24 youth from a separate school through a cognitive pretest. All exploratory modifications were discussed with US survey experts at each step of the adaptation process to ensure fidelity to the original survey. A more detailed description of the translation and adaptation process is provided elsewhere (de Oliveira Corrêa, 2014).

Data collection

Two high schools in São Sebastião were selected for data collection. Recruitment targeted all classes except for adult education and special needs classes. Students were invited to participate in advance and, if under age 18, received a passive consent form to take home to their parents. Data collection took place from November 2013 to February 2014. Students who had permission from their parents were asked to sign an assent form. Instructions were read aloud to students prior to administration. Students who shared answers during data collection were excluded from the sample. Participants were assured of confidentiality of their answers and that they could quit the study without negative consequences. All study procedures were approved prior to data collection by the University of Brasília’s Research Ethics Committee for the Institute of Human Sciences.

Data analysis

Internal consistency was ascertained by calculating coefficient alphas for each scale of the adapted measure. The adopted standard for acceptable reliability was α ≥ 0.60, the same criterion used in the development of the original survey (Arthur et al., 2002). Exploratory modifications were made (i.e. item deletion or inclusion) if they led to a substantial increase in a scale’s coefficient alpha. Confirmatory factor analyses (CFAs) were conducted to identify whether original measurement models fit the data adequately.

We first tested the measurement models for each separate factor. We then assessed the fit of each correlated factor model, run separately for each domain, with attention to factor loadings and inter-factor correlations. The peer-individual domain was analyzed together because its constituting factors are difficult to separate and thus it is often treated as a single domain in the literature (Arthur et al., 2002; Brown et al., 2009). Goodness of fit was assessed by root mean square error of approximation (RMSEA) < 0.06, Comparative Fit Index (CFI) > 0.95, and Tucker–Lewis Index (TLI) > 0.95 (Hu and Bentler, 1999). Adjustments to factor structure at each step were guided by inspection of modification indices. CFAs were conducted using Mplus v.8, with weighted least squares means and variances estimation, which is appropriate for the ordered-categorical nature of the item response options (Asparouhov and Muthén, 2010).

RESULTS

Results of this study are organized into two parts: first, the examination of risk factor scale reliability using coefficient alpha as a measure of scale internal consistency; and second, the examination of scale construct validity using CFA.

Scale reliability

In the community domain, an item (Have you changed schools in the past year?) from the Transitions and Mobility scale was removed due to low correlation with the other items, leading to an improvement in scale internal consistency (from α = 0.57 to 0.62). Similarly, for the peer-individual domain, an item (How many of your best friends have been members of a gang?) from the Gang Involvement scale was correlated weakly with the remaining items. Although removal of this item would have increased the coefficient alpha (from α = 0.80 to 0.85), this modification would not be justifiable. After the modifications, internal consistency for the scales ranged from α = 0.57 to 0.84, with the lowest and only unacceptable levels pertaining to (i) Academic Failure (r = 0.57) and (ii) Rebelliousness (α = 0.57). Internal consistency for the Perceived Availability of Handguns scale was not assessed because it consisted of a single item. Table 1 summarizes the range of coefficient alphas for multiple US studies (Arthur et al., 2002; Brown et al., 2009; SDRG, 2012), as well as for the scales adapted for use in Brazil, after all modifications and adjustments had been performed.

Table 1:

Coefficient alpha for risk factors (with example item), for both the US and the Brazilian samples

Domain	Risk Factor (number of items)	Coefficient	Coefficient
	Example item	Alpha (US)	Alpha (Brazil)
Community	Low Neighborhood Attachment (three items)	0.83–0.87	0.84
	I'd like to get out of my neighborhood.
	Community Disorganization (five items)	0.82–0.85	0.71
	How much do each of the following statements describe your neighborhood? Crime and/or drug selling.
	Transitions and Mobility (four items)	0.64–0.76	0.62
	How many times have you changed homes since kindergarten?
	Perceived Availability of Drugs (four items)	0.75–0.87	0.82
	If you wanted to get some cigarettes, how easy would it be for you to get some?
	Perceived Availability of Guns (1 item)	NA	NA
	If you wanted to get a handgun, how easy would it be for you to get one?
	Laws and Norms Favorable to Drug Use (seven items)	0.75–0.82	0.74
	If a kid smoked a cigarette in your neighborhood would he or she be caught by the police?
School	Academic Failure (two items)a	0.68–0.79	0.57
	Are your school grades better than the grades of most students in your class?
	Low Commitment to School (seven items)	0.65–0.79	0.65
	How interesting are most of your courses to you?
Family	Family History of Antisocial Behavior (10 items)	0.80–0.84	0.80
	Has anyone in your family ever had a severe alcohol or drug problem?
	Poor Family Management (eight items)	0.67–0.86	0.77
	My family has clear rules about alcohol and drug use.
	Family Conflict (three items)	0.62–0.80	0.64
	People in my family have serious arguments
	Parental Attitudes Favorable Towards Drug Use (three items)	0.73–0.83	0.75
	How wrong do your parents feel it would be for you to smoke cigarettes?
	Parental Attitudes Favorable Towards Antisocial Behavior (three items)	0.72–0.82	0.77
	How wrong do your parents feel it would be for you to pick a fight with someone?
Peer-individual	Rebelliousness (three items)	0.69–0.77	0.57
	I ignore rules that get in my way
	Early Initiation of Drug Use (four items)	0.80	0.73
	How old were you when you first had more than a sip of alcohol?
	Early Initiation of Antisocial Behavior (four items)	0.60–0.76	0.63
	How old were you when you first got arrested?
	Gang Involvement (four items)	0.87	0.80
	Have you ever belonged to a gang?
	Favorable Attitudes Towards Drug Use (four items)	0.79–0.89	0.80
	How wrong do you think it is for someone your age to smoke cigarettes?
	Favorable Attitudes Towards Antisocial Behavior (five items)	0.72–0.84	0.71
	How wrong do you think it is for someone your age to take a handgun to school?
	Perceived Risk of Drug Use (four items)	0.82–0.96	0.67
	How much do you think people risk harming themselves if they try marijuana once or twice?
	Friends’ Use of Drugs (four items)	0.79–0.85	0.81
	In the past year, how many of your best friends have used marijuana?
	Interaction with Antisocial Peers (six items)	0.66–0.89	0.81
	In the past year, how many of your best friends have been suspended from school?
	Rewards for Antisocial Involvement (four items)	0.82–0.90	0.76
	What are the chances you would be seen as cool if you carried a handgun?
	Intentions to Use Drugs (three items)	0.65–0.69	0.65
	When I am an adult I will drink beer, wine, or liquor.
	Sensation Seeking (three items)	0.74–0.81	0.76
	How many times have you done what feels good no matter what?

US data from Arthur et al., 2002; Brown et al., 2009; SDRG, 2012.

NA, not applicable (i.e. single-item construct).

^aPearson’s r coefficient provided for two-item factor.

Scale construct validity

Model fit indices for each factor and correlated factor models are presented in Table 2.

Table 2:

Comparison of model fit indices

Factor name	Model	χ 2	df	n	CFI	TLI	RMSEA
Community domain
Community Disorganization	Original	93.31***	5	882	0.956	0.913	0.142
	Modified	9.68*	4	882	0.997	0.993	0.040
Perceived Availability	Original	249.36***	2	891	0.955	0.864	0. 373
	Modified	0.02	1	891	NA	NA	NA
Laws and Norms Favorable to Substance Use	Original	1250.04***	14	892	0.812	0.719	0.315
	Modified	20.60	13	892	0.999	0.998	0.026
Correlated Factor Model	Original	731.25***	192	898	0.966	0.959	0.056
	Modified	444.86***	191	898	0.984	0.980	0.038
Family Domain
Family History of Antisocial Behavior	Original	2013.04***	35	891	0.929	0.909	0.252
	Modified	122.93***	26	891	0.996	0.995	0.065
Poor Family Management	Original	264.15***	20	868	0.918	0.886	0.119
	Modified	159.18***	19	868	0.953	0.931	0.092
Correlated Factor Model	Original	607.85***	283	891	0.989	0.987	0.036
School Domaina
Correlated Factor Model	Original	55.20***	14	907	0.952	0.928	0.057
	Modified	32.50**	13	907	0.977	0.964	0.041
Individual Domain
Gang Involvement	Original	5.05	2	907	0.999	0.998	0.041
Perceived Risk of Drug Use	Original	23.10***	2	893	0.990	0.969	0.109
	Modified	7.65**	1	893	0.997	0.980	0.086
Early Initiation of Drug Use	Original	71.46***	2	899	0.963	0.889	0.197
	Modified	6.32*	1	899	0.997	0.983	0.077
Favorable Attitudes Towards Drug Use	Original	16.46***	2	903	0.996	0.987	0.089
	Modified	0.02	1	903	NA	NA	NA
Favorable Attitudes Towards Antisocial Behavior	Original	30.02***	5	906	0.981	0.962	0.074
Rewards for Antisocial Involvement	Original	2.52	2	905	1.000	0.999	0.017
Friends’ Use of Drugs	Original	9.42**	2	897	0.998	0.993	0.064
Interaction with Antisocial Peers	Original	43.84***	9	894	0.979	0.965	0.066
Correlated Factor Model	Original	2097.97***	965	908	0.945	0.938	0.036
	Modified	1955.22***	964	908	0.952	0.946	0.034

NA, not applicable (Fit indices omitted for modifications that led to just-identified models).

^aCorrelated factor model consisted of a single factor (i.e. Low Commitment to School).

^* p < 0.05;

^** p < 0.01;

^*** p < 0.001.

Community domain

For the community domain, analyses were conducted separately for three risk factors identified by four or more items. Model fit for Community Disorganization was improved by correlating residual errors between two items: Lots of empty or abandoned buildings in my neighborhood and Lots of graffiti in my neighborhood; ΔCFI = 0.041, ΔTLI = 0.080 and ΔRMSEA = −0.102. Similarly, notable modification indices for Perceived Availability of Drugs indicated a potential correlation between two residual errors: How easy would it be for you to get some beer, wine or hard liquor? and How easy would it be for you to get some cigarettes? The inspection of modification indices for Laws and Norms Favorable to Drug Use suggested the separation of this factor into two: (i) Laws Favorable to Drug Use and (ii) Norms Favorable to Drug Use. This modification was supported by inter-item correlations and by previous research (Glaser et al., 2005). A second model was thus tested for two separate but correlated factors, leading to improvements in fit; ΔCFI = 0.187, ΔTLI = 0.279 and ΔRMSEA = −0.289.

The correlated factor model for the community domain was tested last, with 22 items loading on 6 risk factors for this domain (one item was removed previously to improve reliability, Laws and Norms Favorable to Drug Use was treated as two separate scales, and the Perceived Availability of Handguns scale was not included in the analyses). Modification indices suggested cross-loading of one item (I feel safe in my neighborhood) on both the Community Disorganization and Neighborhood Attachment factors, which led to improved model fit; ΔCFI = 0.018, ΔTLI = 0.021 and ΔRMSEA = −0.018. Standardized factor loadings were all statistically significant (ps < 0.05) and ranged from λ = 0.31 to 0.93. Most loadings were good (λs ≥ 0.55; Tabachnick and Fidell, 2007), the only exceptions being an item (Lots of empty or abandoned buildings in my neighborhood) with a factor loading of λ = 0.31 on the Community Disorganization scale, and the item (I feel safe in my neighborhood) that cross-loaded on the Community Disorganization and Neighborhood Attachment scales (λs = 0.43 and 0.41, respectively). Most correlations among factors were statistically significant and ranged from rs = 0.06 to 0.40.

School domain

For the school domain, we observed that the Academic Failure factor did not meet the level of construct validity established in the US studies. Analyses were thus restricted to the Low Commitment to School factor, identified by seven items. Inspection of modification indices suggested correlating residual errors for two items: How often did you enjoy being in school? and How often did you hate being in school? This led to significant improvements in fit; ΔCFI = 0.025, ΔTLI = 0.036 and ΔRMSEA = −0.016. Standardized factor loadings were all statistically significant yet substandard (λs < 0.45; Tabachnick and Fidell, 2007), with the exception of one item (How interesting are most of your courses to you?), which had a factor loading of λ = 0.65.

Family domain

For the family domain, we initially conducted the analysis with the only two factors identified by more than three items. One item (Has anyone in your family ever had a severe alcohol or drug problem?) loaded poorly (λ = 0.22) on the Family History of Antisocial Behavior scale and was removed. Further, inspection of modification indices and inter-item correlations recommended separating the factor into two parts: (i) Sibling History of Antisocial Behavior and (ii) Adult History of Antisocial Behavior. We assessed a second model that correlated the two separate factors and identified improved fit; ΔCFI = 0.067, ΔTLI = 0.086 and ΔRMSEA = −0.187. Model fit also was assessed for Poor Family Management, with fit being improved after correlating residuals for two items: The rules in my family are clear and My family has clear rules about alcohol and drug use; ΔCFI = 0.035, ΔTLI = 0.045 and ΔRMSEA = −0.027.

The correlated factor model for the family domain was assessed for 26 original items loading on six risk factors (four factors theorized for this domain and another factor split into two). Standardized factor loadings were all statistically significant, ranging from λs = 0.43 to 0.98. Most loadings were excellent (λs ≥ 0.71; Tabachnick and Fidell, 2007) and many were good, except for two items on the Poor Family Management scale: My parents ask if I've gotten my homework done and Would your parents know if you did not come home on time? (λs = 0.43 and 0.53, respectively. Correlations were all statistically significant and ranged from rs = 0.11 to 0.87. The largest correlation was between (i) Parental Attitudes Favorable Toward Drug Use and (ii) Parental Attitudes Favorable to Antisocial Behavior (r = 0.87). This strong correlation suggested that both scales could be condensed into one general Parental Attitudes factor, but we chose to keep the two scales separated and preserve their utility in distinguishing between drug-specific and non-drug-specific attitudes (i.e. different communities may have different prevention interests and targets).

Peer-individual domain

For the peer-individual domain, separate analyses were conducted for nine risk factors, each identified by four or more items. Model fit indices for four of these factors met our criteria and thus warranted no further exploratory modifications: Gang Involvement, Rewards for Antisocial Involvement, Friends’ Use of Drugs and Interaction with Antisocial Peers. We observed an issue with an item (How old were you when you first got arrested?) in the Early Initiation of Antisocial Behavior scale. Almost all participants chose the same answer for that question, which indicated a need to remove it.

Model fit for Perceived Risk of Drug Use was improved slightly by correlating residuals between two items about marijuana: How much do you think people risk harming themselves if they try marijuana once or twice? and How much do you think people risk harming themselves if they smoke marijuana regularly?; ΔCFI = 0.007, ΔTLI = 0.011 and ΔRMSEA = −0.023. Similarly, model fit for Early Initiation of Drug Use was improved by correlating residuals for two items about alcohol consumption: How old were you when you first had more than a sip of alcohol? and How old were you when you first began drinking alcoholic beverages regularly?; ΔCFI = 0.034, ΔTLI = 0.094 and ΔRMSEA = −0.120. Inspection of modifications indices for Favorable Attitudes Towards Drug Use suggested a potential correlation between two residual errors: How wrong do you think it is for someone your age to drink alcohol regularly? and How wrong do you think it is for someone your age to smoke cigarettes? Modification indices for Favorable Attitudes Towards Antisocial Behavior did not suggest a clear or justifiable path, and thus no adjustments were made.

The correlated factor model for the peer-individual domain included 47 original items pertaining to 12 theorized risk factors. Inspection of modification indices suggested cross-loading one item (How many of your best friends have been members of a gang?) from the Gang Involvement scale on the Interaction with Antisocial Peers factor, leading to improved model fit; ΔCFI = 0.007, ΔTLI = 0.008 and ΔRMSEA = −0.002. Standardized factor loadings were all statistically significant and ranged from λs = 0.19 to 0.98. Most loadings were excellent or at least good, with two exceptions: (i) How much do you think people risk harming themselves if they smoke one or more packs of cigarettes per day?, which loaded poorly (λ = 0.42) on the Perceived Risks of Drug Use scale, and (ii), the item that cross-loaded on the Gang Involvement and Interaction with Antisocial Peers scales (λ = 0.19 and λ = 0.72, respectively). Correlations were all significant and ranged from rs = 0.29 to 0.77.

DISCUSSION

Results of this study provide a foundation for the use of an adapted version of the CTCYS to assess RPFs for drug use and violence in Brazilian populations. The original US survey was translated into Brazilian Portuguese and adapted for use in Brazil following a rigorous process of independent translations of the measure, synthesis of the translations, consulting field experts, and testing the measure with the target population (Borsa et al., 2012). All changes were discussed with survey experts from among the original CTCYS developers, who helped ensure that core features remained unchanged. The adapted survey was administered to high school students in a Brazilian community, and the data were used to calculate scale reliability and model fit for each domain of the survey.

Scale reliability

Internal consistency analyses indicated the need for some adjustments. In the community domain, an item assessing having changed schools in the past year had very low correlations with the other items in the Transitions and Mobility scale, possibly due to many students having changed schools in the transition from middle to high school (62.62% of participants who answered Yes were ninth graders). Removing this item led to improved scale reliability, after which all community scales met reliability criteria. Reliability levels for this Brazilian sample were similar to the US studies of these scales (Arthur et al., 2002; Brown et al., 2009; SDRG, 2012), with the exception of Community Disorganization.

For the school domain, no exploratory modifications were possible to improve the reliability (r = 0.57) of the Academic Failure scale, which was substantially lower than in the US studies (rs = 0.68–0.79).

In the family domain, reliability analyses indicated acceptable internal consistency for all risk scales and were comparable with values of coefficient alpha found in the US studies.

For the peer-individual domain, we observed that removing the item indicating friends who were gang members would improve the reliability of the Gang Involvement scale; however, we rejected this modification given lack of rationale for removing the item. We similarly rejected modifications that would improve the Rebelliousness scale’s reliability based on lack of justification, which resulted in a coefficient alpha (α = 0.57) somewhat lower than in other studies (αs = 0.69–0.77). Another notable difference in reliability levels for Brazil and the US was observed for Perceived Risk of Drug Use scale (αs = 0.67 vs. 0.82–0.96, respectively). Scale reliability for all other factors in this domain was generally similar to findings from US studies.

Overall, Rebelliousness and Academic Failure scales had the weakest transferability from US to Brazilian surveys. In the case of the Rebelliousness scale, this could be a consequence of different perceptions of what rebellious behavior is, where deviant behavior in the US could be considered adaptive in Brazil (Ferreira et al, 2012). For Academic Failure, the issue could be rooted in differences regarding conceptualizations of academic performance (Esteban, 2009; Villani and Oliveira, 2018).

Reliability levels were generally similar between the data collected from our Brazilian sample and data collected in Iran (Baheiraei et al., 2016), with the exception of (i) Low Neighborhood Attachment, (ii) Sensation Seeking, (iii) Low Commitment to School and (iv) Perceived Risk of Drug Use (coefficients alphas for the first two scales were higher for Brazil, while coefficients for the last two were higher for Iran). When compared with the Malaysian study (Razali and Kliewer, 2015) reliability levels for the Brazilian sample were lower. However, it is noteworthy that the coefficient alphas for Malaysia were generally higher than those originally observed in the US.

Scale construct validity

CFA suggested that the correlated factor models generally fit the data well; however, some factors had substandard fit and warranted exploratory modifications. A notable example was observed in the community domain; model fit initially was not good but improved substantially after the Laws and Norms Favorable to Drug Use scale was split into two factors, one for Laws and one for Norms. A possible explanation is that police action to repress youth drug use and antisocial behavior does not necessarily reflect on residents’ acceptance of youth drug use, and thus the items would be measuring two different constructs (Glaser et al., 2005). By inspecting the inter-item correlation matrix, we also identified stronger correlations among items referring to Norms and items referring to Laws, and weaker correlations between these types of items. Such distinction between Laws and Norms also was observed by researchers in Iran when validating their adaption of the CTCYS (Baheiraei et al., 2016).

Justifiable post-hoc correlation of item residual errors also was indicated to improve model fit. Thus, we correlated residual errors for youth report of empty buildings and graffiti in their neighborhood, understanding that both items might be measuring urban blight. Residuals for items on perceived availability of alcohol and of cigarettes also were allowed to be correlated, considering that both are licit drugs and perceived differently in the community when compared with illicit drugs. Modification indices indicated the potential to improve model fit by cross-loading ‘I feel safe in my neighborhood’ on both the Community Disorganization and the Neighborhood Attachment factors. This suggests some item ambiguity that calls for greater attention to discriminant validity in future adaptations of these measures.

In the school domain, the correlation of residual errors between items about hating school and enjoying school suggested that, to some extent, these items might have an artifactual relationship that is not captured by the underlying construct.

In the family domain, Family History of Antisocial Behavior had to be separated into two different factors, one for Sibling History and another for Adult Family Member History. Although these findings are not corroborated by previous studies, it is not surprising that youth might report different perceptions and problem behavior rates for adult family members and their siblings (who may even be in their own age range). Additionally, we had to remove an item in the Family History of Antisocial Behavior factor, reporting on family history of severe drug use problems, due to a low factor loading. This item stood out when compared with other items in the scale, not only because it was a dichotomous question but also due to its subjectivity when compared with the other, more objective items. Furthermore, the correlated residuals between an item about having clear family rules in general and another on clear family rules about drug use suggests that these are not distinct representations of the underlying construct, and that youth may just be lumping them together conceptually.

In the peer-individual domain, good fit to the data was observed for four risk factors, which had good or excellent model fit and warranted no modifications: Gang Involvement, Rewards for Antisocial Involvement, Friends’ Use of Drugs, and Interaction with Antisocial Peers. Nonetheless, a case warranting item removal was observed in the Early Initiation of Antisocial Behavior factor, regarding an item about how old participants were when they were first arrested. Looking at the proportion of answers we observed that 99.10% of students reported never being arrested, bringing the item’s utility into question. This could be evidence that more antisocial youth had already dropped out of school before ninth grade and, as such, were not represented in our analyses.

We allowed residual errors for items about individual attitudes towards alcohol use and cigarette use to be correlated, since these drugs are legal in the adult population and possibly have a differential perception of harm vis-à-vis illegal drugs. We also correlated residual errors for two items about perceived harm of sporadic and of regular marijuana use. The correlation between these items may represent an artifactual relationship arising from the commonality of marijuana in both items. Similarly, we correlated residual errors for two items reporting age of first use of alcohol and age of onset of regular drinking, which, as with the previous example, also may represent an artifactual relationship due to the shared focus on a single drug. Moreover, modification indices suggested a potential improvement to model fit through cross-loading an item about friends who are gang members on the Gang Involvement and Interaction with Antisocial Peers factors. Similar to what we observed for the community domain, this suggests some ambiguity in this item and warrants further investigation of discriminant validity for these adapted measures.

Notwithstanding these exploratory modifications, results suggested that, in general, measurement models for our pilot Brazilian data replicated the reliability and construct validity of the original US risk factor measures. Conversely, CFAs indicated good model fit for most risk factors in the community, school, family and peer-individual domains, after justifiable modifications (e.g. correlating residual variance for items with common wording or foci) were made. Nonetheless, six factor models did not fit the data even after modifications: Poor Family Management, Family History of Antisocial Behavior, Perceived Risk of Drug Use, Early Initiation of Drug Use, Favorable Attitudes Towards Antisocial Behavior, and Interaction with Antisocial Peers. These poorly fitting factors present a challenge for future research aiming at the further refinement of this instrument for use in Brazil. Low factor loadings for the school domain and marginal fit for the correlated factor model in the peer-individual domain also suggest a need for further adaptations of these scales.

This study had some limitations. As an exploratory pilot study, the data were collected from a single community in Brazil, and we do not expect it to be necessarily representative of the broad racial, cultural, and societal diversity of the country. Generalizability of results is also limited by lower-than-expected participation rates. Additionally, only high school students were recruited, and it is possible that we did not have access to high-risk youth with higher drug use and antisocial behavior rates, who by ninth grade may have already dropped out of school. Data collection also was subject to school capacity and staff availability, and thus occurred under varying conditions, such as some data being collected in an auditorium and some in classrooms, or teachers being present for some sessions but not for others. As a pilot experimental study of the psychometric properties of risk factor scales in a Brazilian community, it is paramount that future studies be conducted to test the replicability of our findings and advance the utility of the CTCYS risk factors.

Despite the limitations, this study represents an important step in youth drug use and violence prevention research in Brazil. The translated risk scales for the community, school, family, and peer-individual domains generally have shown acceptable reliability and good fit. CTCYS scales hold great potential to assist decision makers in identifying which risk factors are more prevalent and what interventions are more appropriate to address them, as well as choosing new interventions to translate and adapt for use in the country. These data also may be used to monitor the impact of existing interventions, policies, and strategies on risk factors, which function as proximal outcomes in the prevention of youth problem behaviors. Furthermore, community leaders can use the CTCYS to develop profiles for their own communities, helping them advocate for government support in meeting their youth development needs.

Future studies should seek to validate this adapted measure with a new sample of Brazilian youth. Researchers should also focus on the associations between risk factors and youth outcomes, assessing reliability and validity for protective factor scales, as well as collecting data from larger, more generalizable samples, both from middle school students and residents of other Brazilian communities. Broader data will allow for a better understanding of the functioning or need for improvement of risk factor scales in these major risk domains.