We conducted a school-based survey within a prevention-in-action process. A group of researchers developed the survey in Iceland (Sigfúsdóttir et al., 2009; Kristjansson et al., 2020a,b). In 2018, the implementation of this community prevention model of substance use in adolescents was initiated in six municipalities of the Metropolitan Region of Greater Santiago in Chile based on a collaboration between the Icelandic Centre for Social Research and Analysis (ICSRA) and the Universidad de Chile. The prevention action included a survey of tenth-grade students conducted every 2 years to base the community prevention work on local data and evaluate the prevention process.

The prevention teams of the municipalities invited schools in their localities to participate. When the schools agreed to participate, written information was sent to the parents. Municipal prevention teams coordinated and supervised implementation of the survey in the schools, with the support and guidance of the Universidad de Chile, following the protocols developed by ICSRA (Kristjansson et al., 2013). We translated the English version of the survey provided by ICSRA into Spanish and adapted the linguistic suitability and content for the context in Chile (Supplementary material 1). This process involved an expert panel and included a semantic validation and assessment of understanding through a pilot study at an adolescent healthcare center. The pencil-and-paper surveys applied on-site were administered in all schools in the same week of June 2018 (Libuy et al., 2023). The surveys were applied during class hours inside the classroom, with the teacher present in the room. The responses were confidential and were not observed by the teacher. The answers were not shared among classmates either. The students had enough time (up to 60 min) to answer the survey. The process of applying the survey, its follow-up, the characteristics of the communities, and the prevention process carried out have also been addressed in previous publications (Contreras et al., 2022; Libuy et al., 2023).

After collecting the completed surveys from the schools, they were scanned and sent to ICSRA, which built the database. We removed duplicates and cases with less than 50% of the responses completed.

The participation of the municipalities, schools, and students was voluntary, using a passive informed consent procedure for the parents and an assent form for the students. The questionnaire was anonymous, protecting the identity of the students. The anonymized data were managed and stored by the research team. The study was approved by the ethics committee of the Hospital Clínico Universidad de Chile (OAIC 981/18).

The survey assessed the prevalence of substance use and several protective and risk factors (see Supplementary material). The protective factors included parental support, monitoring and communication, school wellness, and participation in extracurricular activities such as organized sports activities. Among the risk factors were unsupervised leisure time, lifestyle, and substance use among peers (Kristjansson et al., 2013). Outcome variables were the proportion of adolescents with lifetime alcohol use, past-month alcohol use, lifetime drunkenness, past-month drunkenness, lifetime cannabis use, and cannabis use in life ranging from 1 to 40 times or more (categories: 1 to 2 times, 3 to 5, 6 to 9, 10 to 19, 20 to 39, and 40 or more times in life). The independent variable was the score on the parenting scale. The parenting scale comprised 16 items scored on a Likert-like scale from 1 to 4 by the students. Higher scores indicated higher levels of parental support. For the first 5 items, the initial question was “How easy or hard would it be for you to receive the following from your parents?” and the items were as follows: 1. Caring and warmth, 2. Discussions about personal affairs, 3. Advice about studies, 4. Advice about other issues (projects) of yours, and 5. Assistance with things. For items 1 to 5, a score from 1 to 4 was assigned to answer the alternatives: very difficult, rather difficult, rather easy, or very easy. The introductory question to items 6 to 16 was “How do the following statements apply to you?” and the items were as follows: 6. My parents find it important that I do well in my studies, 7. My parents set definite rules about what I can do at home, 8. My parents set definite rules about what I can do outside the home, 9. My parents set definite rules about when I should be home in the evening, 10. My parents know with whom I am in the evenings, 11. My parents know where I am in the evenings, 12. My parents know my friends, 13. My parents know the parents of my friends, 14. My parents often talk to the parents of my friends, 15. My parents and the parents of my friends sometimes meet to talk to one another, and 16. My parents follow what I do in my recreational time. For items 6 to 16, a score from 1 to 4 was assigned to answer the alternatives: applies very poorly to me, applies rather poorly to me, applies rather well to me, or applies very well to me.

Item analysis of the scale was performed, describing the rate of missing responses, mean scores for each item, standard deviation, skew, item difficulty (rate of ideal response), item discrimination, and Cronbach’s alpha if the item was deleted. Reliability for the scale was analyzed using Cronbach’s alpha and calculating the omega. In the item analysis, total parenting scores were treated as a continuous variable. The Kaiser-Meyer-Olkin factor was calculated, and Bartlett’s K-squared before factor analysis. Exploratory factor analysis was run using polychoric matrices, and the factors were retained with Kaiser’s rule of Eigenvalues over one.

The six municipalities were compared by measuring the invariance using multiple-group confirmatory factor analysis. The changes in model fit indices, such as comparative fit index (CFI) and root-mean-square error of approximation (RMSEA), were assessed for measurement invariance by comparing changes between fit indices at baseline and in equality constraint models (thresholds and thresholds and loadings). For testing loading invariance, a change of ≥ −0.010 in CFI, together with a change of ≥0.015 in RMSEA, indicated non-invariance (Svetina et al., 2020).

Multilevel logistic regression models were performed to analyze the association between the outcomes (lifetime alcohol use, past-month alcohol use, lifetime drunkenness, past-month drunkenness, lifetime cannabis use, and cannabis use in life ranging from 1 to 40 or more times, in six dichotomic categories) and the total parenting score. The outcomes were dichotomic, and the reference category was no substance use. In the multilevel logistic regression, the score of the parenting scale was dichotomized to facilitate the interpretation of the results: high (over the median) and low (equal or under the median). We included the following control variables: gender (categorical dichotomous), age (continuous), living with both parents (categorical nominal), educational level of the parents (categorical nominal), employment status of the parents (categorical nominal), friends using substances (categorical dichotomous), school funding (categorical nominal), and municipality (categorical nominal). Adjusted odds ratios (AOR) were calculated for each outcome. In addition, each subscale of parenting was also dichotomized between high (over the median) and low (equal or under the median), and adjusted odds ratios (AOR) were calculated for each outcome. Multicollinearity between independent variables was assessed with the variance inflation factor (VIF). The VIF values were under five. Thus, collinearity was not present. The linear relationship between independent variables and the logit of the outcome was assessed using scatter plots. In addition, no influential values or outliers were observed in the data.

In the present study, we used a multilevel model to control the cluster effects of the schools and municipalities in which the students were from. The multilevel model allows for analyzing data that is not randomly distributed but grouped in clusters. Intra-class correlations were calculated to estimate the proportion of variance explained at the school and municipality levels. Finally, interactions between gender and parenting were explored using multilevel logistic regression models for lifetime alcohol use, past-month alcohol use, lifetime drunkenness, past-month drunkenness, lifetime cannabis use, and cannabis use in life of 10 or more times. The fitted values of the parenting scores were plotted by gender.

The analyses were performed with the software R version 4.0.1 using the packages psych for psychometrics (Revelle, 2023), items, and factor analyses [fa.poly and omega (Zinbarg et al., 2005) functions of the psych package]. Packages lavaan and semTools for software R were used to measure invariance (Svetina et al., 2020). The packages lme4 (Bates et al., 2015) and insight (Lüdecke et al., 2019) were used for multilevel regressions and intra-class correlation, and sjPlot for plot interactions in models. Raw data are available from the corresponding author upon request.

A total of 7,538 tenth-grade students from 118 schools and six municipalities were included. The response rate was 86.9%. The mean age of the participants was 16.0 (SD = 0.7) years; 48.7% were girls and 51.3% were boys. Table 1 summarizes the description of the sample.

The 16 items of the parenting scale had a mean response rate of 98.5%; on average, 77% of the answers were the expected ideal alternative (score = 4). Table 2 shows response rates, item discrimination, and the expected response rates for each item (also called item difficulty). The mean sum score was 49.1 (SD = 7.5), and the median was 50. The total scale of 16 items showed good internal consistency with a Cronbach’s alpha of 0.84 (95% CI: 0.83–0.85). The alpha, if items were deleted, was ≤0.84 for each item. The omega total was 0.84, and the omega hierarchical was 0.54.

The Kaiser-Meyer-Olkin factor of 0.84 and Bartlett’s K-squared = 4283.3 (df = 15; p < 0.001) indicated appropriateness for the factor analysis. Exploratory analysis retained three factors. Factor 1: Items 1 to 5; Factor 2: items 6 to 11 and item 16; and Factor 3: items 12 to 15, with eigenvalues of 1.6, 1.5, and 1.5, respectively. Factor 1 was the Relationship between parents and adolescents. Factor 2 reflected parental Norms and monitoring. Factor 3 was Parents knowing their children’s friends and the parents of their children’s friends. Factor 1 showed high loadings, with unweighted least squares (ULS) over 0.73; Factor 2 had loadings between 0.46 and 0.78; and Factor 3 had loadings between 0.55 and 0.93 (Table 3). RMSEA index was 0.144, and the Tucker-Lewis index of factoring reliability was 0.723. The proportion of the explained variance for each factor was 36% for Factor 1, 34% for Factor 2, and 30% for Factor 3. The internal consistency for each subscale was a total of omega 0.84 for Factor 1, omega 0.77 for Factor 2, and omega 0.81 for Factor 3; and omega hierarchical 0.34 for Factor 1, 0.30 for Factor 2, and 0.26 for Factor 3.

Low inter-factor correlations were observed between Factor 1 and Factor 2 (0.39), between Factor 1 and Factor 3 (0.33), and between Factor 2 and Factor 3 (0.35).

The measurement invariance results between municipalities indicated changes under 0.01 from baseline CFI (0.923) compared to thresholds (0.919) and thresholds and loadings (0.922) in equality constraints models; and additionally, RMSEA indicated changes under 0.015 from baseline (0.093) compared to thresholds (0.089), and thresholds and loadings (0.084) in equality constraints models.

The total parenting score reported was associated with all outcomes of alcohol use: lifetime alcohol use, past month alcohol use, and lifetime and past month drunkenness. Higher scores of parenting on each subscale were associated with lower probabilities of alcohol use. Table 4 reports the adjusted odds ratios (AORs) for parenting. Among all alcohol use outcomes, lifetime alcohol use had the strongest association. Regarding the different aspects of parenting, the subscale Norms and monitoring had the strongest association with alcohol use. The intra-class correlation coefficients observed for the alcohol use outcomes were between 5 and 7% at the school level and lower at the municipality level (see Table 4).

Higher parenting scores were associated with lower odds of cannabis use. For lifetime cannabis use, the AOR of a high total parenting score was 0.54 (95% CI: 0.47–0.61); and the AOR of the subscale Relationship between parents and adolescents was 0.69 (95% CI: 0.70–0.90); the AOR of the subscale Norms and monitoring was 0.50 (95% CI: 0.44–0.57); and the subscale Parents knowing their children’s friends and the parents of their children’s friends was 0.57 (95% CI: 0.50–0.65). Table 5 presents AORs for the total parenting score and subscales with outcomes representing different frequency categories of lifetime cannabis use. For any lifetime cannabis use, high parenting scores had an AOR similar to the alcohol use outcomes. However, parenting had lower AORs for more frequent lifetime cannabis use (see Table 5). All parenting subscales were significantly associated with cannabis use in any frequency, similar to the total parenting score, except for the subscale Relationship between parents and adolescents and cannabis use in life ranging from 1 to 2 times. The subscale Norms and monitoring had the strongest association with all categories of cannabis use frequencies (Table 5).

Intra-class correlation coefficients observed in lifetime cannabis use were 15.8% at the school level and 1.4% at the municipality level. We observed higher intra-class correlation coefficients at the school level in the categories indicating more frequent lifetime cannabis use (Table 5). The intra-class coefficients at the municipality level were lower than those at the school level for the cannabis use outcomes. The school level explained more variance of cannabis use than alcohol use.

The association of high parenting scores with less adolescent alcohol and cannabis use was significantly stronger for girls than boys. In multilevel regression models, significant interactions between parenting and gender were observed, with a greater effect of parenting on substance use in girls than in boys for lifetime alcohol use (p = 0.01), past-month alcohol use (p = 0.005), lifetime cannabis use (p = 0.005), and cannabis use in life of more than 10 times (p = 0.021). However, the interaction was not significant for lifetime drunkenness (p = 0.235) and drunkenness in the past month (p = 0.224). Figure 1 shows the interactions between parenting and gender for substance use outcomes.

This study presents the psychometric properties of a 16-item scale to assess perceived parental practices in tenth-grade secondary school students in Chile. The scale is part of a larger questionnaire on risk and protective factors of adolescent substance use. It showed good internal consistency and a three-factor structure in Chile: 1. Relationship between parents and adolescents; 2. Norms and monitoring; and 3. Parents knowing their children’s friends and the parents of their children’s friends. Parental practices were associated with alcohol and cannabis use among adolescents. The items related to Norms and monitoring had the strongest association with substance use. Parental practices showed a stronger association with substance use in girls than in boys.

Few parenting instruments have strong psychometric properties, and very few have been tested in the Spanish language. In Mexican-American populations, cross-language measurement equivalence was investigated for warmth, consistent discipline, harsh parenting, and parental monitoring, showing reliability between Cronbach’s alpha 0.7 to 0.8 for each scale and adequate convergent construct validity equivalence across languages (Nair et al., 2009). The Parenting Children and Adolescents (PARCA) scale was nearly invariant in the Latino population in the United States in the Spanish version, showing a reliability of Cronbach’s alpha 0.95, and the authors described a confirmatory factor analysis with good model fit for a three-factor structure: supporting good behavior, proactive parenting, and setting limits (Ringle et al., 2019). Outside the US, perceived parental monitoring was assessed in Mexico City with a scale adapted from Kerr and Stattin that showed good reliability with a Cronbach’s alpha of 0.82 (Stattin and Kerr, 2000; Kerr et al., 2010), similar to the reliability of the scale used in the present study. In Colombia, however, based on the Communities That Care survey, the measure of attitudes of parents towards adolescent drug use had limited reliability with a Cronbach’s alpha coefficient of 0.57 (Mejía-Trujillo et al., 2015).

Parenting scores were associated with all substance use outcomes in our study, indicating construct validity. The odds ratios reported in this study had small to medium effect sizes (Chen et al., 2010). A meta-analysis of longitudinal studies on alcohol use in adolescents reported a small pooled effect size for the predictors: parent–child relationship quality, parental involvement, parental monitoring, and parental support, with correlation coefficients between −0.1 and − 0.2 (Yap et al., 2017). In line with our results, a higher effect size was reported for parental monitoring. Regarding cannabis, it has been reported that general support from parents and parents knowing their children’s friends were preventive for lifetime use (OR 0.72 and 0.27, respectively) and for past month use (OR 0.68 and 0.21, respectively) (Vermeulen-Smit et al., 2015). Cannabis-specific rules were preventive for cannabis use in the past month with an OR of 0.6. International literature is consistent with our findings about larger effect sizes of parental practices on cannabis compared to alcohol use.

Interestingly, the present study showed that the association of parenting with cannabis use was stronger than the association of parenting with alcohol use; and more frequent cannabis use was more strongly associated with parenting, suggesting a dose-dependent relationship. A high score on the parenting scale was more protective for more severe substance use. Differential effects of parenting on alcohol and cannabis use in adolescents could be related to the legal status of the substances. The use of cannabis in adolescents may be considered more problematic than the use of alcohol. Although legalization is debated and its use decriminalized, cannabis is still an illegal drug for recreational use in Chile and is less normalized than alcohol (Mundt et al., 2023).

The cluster effect of the schools on substance use, which may also reflect the effect of peers, was higher for cannabis than for alcohol use and especially high for frequent cannabis use. Previous research from Chile also found strong effects of the school environment on cannabis use, compared to alcohol use (Gaete and Araya, 2017; Libuy et al., 2020). The use of cannabis among friends was identified as the most important risk factor for adolescent substance use in Chile (Libuy et al., 2020). However, in line with the theory of normalization, a higher prevalence of cannabis use showed a reduction in the association between cannabis use and cannabis use among friends (Sznitman et al., 2013). The impact of peers on alcohol use in adolescents was lower than on cannabis use (Mason et al., 2017). Parental practices may moderate the effect friends may have on substance use in adolescents (Kiesner et al., 2010; Boyd-Ball et al., 2014).

Our study showed that gender significantly interacted with the association of parenting and substance use of adolescents in Chile; higher parenting scores were more protective for girls than boys regarding alcohol and cannabis use. The male gender role has been linked to more alcohol use and stronger expression of the female role to less alcohol use (Seedat et al., 2009; Shakya et al., 2019). Lower perceived parental monitoring and weaker perceived family relationships were more strongly associated with drinking profiles among females than among males (Strunin et al., 2015). The effect of parental monitoring on criminal conduct was also stronger for girls than for boys (Kerr et al., 2010). Parents appeared to protect girls more than boys. Connectedness to parents and parental control also had a positive effect on risk behaviors in girls. Adolescent disclosure to parents had a stronger association with criminal conduct in girls than in boys (Kapetanovic et al., 2019). Further, in Iceland, girls received more support and monitoring from their parents compared to boys (Kristjánsson and Sigfúsdóttir, 2009).

This is the first validation of the parenting scale administered as part of Planet Youth’s Icelandic Prevention Model in Spanish and Latin America. We used a large sample of adolescents from many schools and diverse municipalities in Chile.

The cross-sectional study design has limitations regarding the inference of causality. The reverse causality or bidirectionality of influences is also possible. Further limitations arise from the self-report survey design without direct assessment of the parents. The adolescents perceived the parenting practice. We did not assess careless or insufficient effort responding, which may affect the study’s internal validity and lead to a response bias. In future studies, cleaning the data with a careless response analysis could help to detect bias, ensure greater internal validity, and determine whether realistic and accurate data are available (Merino-Soto et al., 2021; Alarcon and Lee, 2022). Lastly, only six municipalities were studied. Therefore, the estimate of the effect at the municipal level may not be accurate.