- Email: [email protected]
Predictive Model of Weapon Carrying Among Urban High School Students:
JOURNAL OF ADOLESCENT HEALTH 1998;22:312–319
ORIGINAL ARTICLE
Predictive Model of Weapon Carrying Among Urban High School Students: Results and Validation JOHN KULIG, M.D., M.P.H., JEANETTE VALENTINE, Ph.D., JOHN GRIFFITH, Ph.D., AND ROBIN RUTHAZER, M.P.H.
Objectives: The purpose of this study was to identify the behavioral, psychosocial, and demographic predictors of self-reported weapon carrying among secondary school students who attend urban public schools. Methods: Self-reported weapon carrying was measured in a schoolwide anonymous health survey conducted in two demographically comparable high schools in 1992, in Boston, Massachusetts. Indicators of self-perception, depression, stressful life events, and adolescent risk behaviors of substance use and sexual behavior, along with self-reported weapon carrying, were measured. The students in both schools were racially heterogeneous, with the majority of about 80% from black or Hispanic backgrounds. A predictive model was developed using a forward stepwise logistic regression model in one innercity high school, and tested in a second high school. Results: Self-reported lifetime weapon carrying was 32% overall. The major predictors of weapon carrying among urban secondary school students are a combination of demographic, psychosocial, behavioral, and school-related factors. This analysis indicates consistency in eight markers predictive of weapon carrying: lower age, male gender, regular marijuana use, sexual experience, having witnessed a crime, having skipped school, suicidal ideation, and having hit or “beat up” someone.
From the Department of Pediatrics, Division of General Pediatrics and Adolescent Medicine (J.K., J.V.), The New England Health and Poverty Action Center (J.V.) Boston, Massachusetts; and the Department of Internal Medicine, Division of Clinical Care Research (J.G., R.R.), New England Medical Center, Boston, Massachusetts Address reprint requests to: John Kulig, M.D., M.P.H., Director of Adolescent Medicine, New England Medical Center, 750 Washington Street, Box 479, Boston, MA 02111. Manuscript accepted July 11, 1997. 1054-139X/98/$19.50 PII S1054(98)00256-5
Race parental education, and family composition were not significant predictors. Significant predictors of weapon carrying were marijuana use and sexual experience, each of which was consistently high in both schools. Conclusions: The model-building and validation presented in this study provide empirical evidence for three important conclusions. First, weapon carrying is associated with multiple and interrelated factors which include demographic, psychosocial, behavioral, and school-related characteristics of high school–age adolescents. Second, students with more risk factors are more likely to carry a weapon, suggesting that the variables are independent markers. Third, this study identified marijuana use and being sexually experienced as both highly predictive of weapon carrying. Implications of this study for prevention point to the need for comprehensive multidisciplinary services in high school that include mental health counseling as well as health education efforts aimed at behavior change. © Society for Adolescent Medicine, 1998 KEY WORDS: Violence Weapon(s) Students High school Survey
Weapon carrying among youth is a significant risk factor associated with adolescent morbidity and mortality. Homicide remains the second leading cause of mortality in the U.S. population ages 15–24,
© Society for Adolescent Medicine, 1998 Published by Elsevier Science Inc., 655 Avenue of the Americas, New York, NY 10010
April 1998
accounting for 8019 deaths in 1992 (1). From 1979 to 1992, the age-specific death rate from homicide rose from 14.2 to 22.2/100,000 population in this age group (1). The rate of 37.7 among males is six times that of the female rate of 6.4, and homicide is the most common cause of death among black males age 15–24, with the rate of 154.4/100,000 population being three times that of deaths owing to motor vehicle and other accidents (49.9) (1). The availability of a weapon in a violent confrontation significantly increases the risk of a lethal outcome. In 1989, 80% of homicide victims aged 10 –19 were killed with guns, and 10% were stabbed to death (2). Based upon present trends, it has been projected that by the year 2003, firearms will surpass motor vehicles as the leading cause of injury-related death in the United States (3). Understanding who carries weapons and why can provide information of value in preventing and reducing the use of weapons among young people. Healthy People 2000 has established a goal of reducing by 20% weapon carrying among adolescents aged 14 –17 (4), estimated to be 22.1% in 1993 from the annual Youth Risk Behavior Surveillance System of the Centers for Disease Control and Prevention (5). Factors associated with weapon carrying have been the subject of numerous investigations aimed at identifying the causes and correlates of this lifethreatening risk behavior. Empirical evidence about the correlates of weapon carrying suggests that carrying weapons is indicative of and associated with violent and aggressive behavior and is used for purposeful criminal activity, rather than self-protection (6,7). Studies analyzing data from the national Youth Risk Behavior Survey find that weapon carrying is intercorrelated with other adolescent risk behaviors, including fighting that involves injuries, substance abuse, school truancy, and sexual activity, thus suggestive of a syndrome of antisocial behavior among troubled youth (5,8 –11). Gender differences in the prevalence of weapon carrying and the reasons for its use are also important considerations. For males, weapon carrying is likely to be associated with violent aggressive behavior in their community, while for females, weapon carrying is likely to be associated with gang fighting (12). A number of studies of weapon carrying among youth are observational cross-sectional ones which rely on correlational or multivariate analysis that is often limited to co-occurring behavioral measures. Information on social and psychological factors is of critical importance to understanding who carries weapons and why. This information can contribute
WEAPON CARRYING
313
to developing preventive interventions for youth at risk of carrying and using weapons. A developmental perspective on adolescent behavior points to the importance of psychological state and social and familial factors that can help explain antisocial and self-harming behaviors of youth (13–15). The purpose of this study was to go beyond correlational or predictive modeling of interrelated behaviors and to identify the association of both behavioral and psychosocial factors with weapon carrying among high school students. The main study question was: What demographic, psychosocial, and behavioral factors are predictive of self-reported weapon carrying among adolescents attending inner-city, urban secondary schools? The study approach develops a predictive model of weapon carrying from two crosssectional high school student surveys that include adolescent risk behaviors identified in the literature as associated with weapon carrying as well as psychosocial and demographic measures.
Methods Study Design Cross-sectional, schoolwide anonymous surveys were conducted in two secondary schools in Boston in the Spring of 1992. The two schools were chosen based upon willingness of school administrators to participate in a schoolwide health survey and demographic comparability. The surveys were carried out by trained survey administrators. The prearranged administration of the survey was accomplished during a class period of 45 min, with several survey administrators going to multiple classrooms simultaneously on the same day. The questionnaires were distributed in each classroom by the survey administrators who responded to questions when needed and collected the completed surveys. Teachers did not participate in survey administration but remained in the classrooms. The survey was a population-based survey with all students in attendance on the day of survey eligible for participation. The size of the student body at each of the two schools differed. The student body at School A was composed of 975 students. The student body at School B was composed of 1500 students. Indicators of selfperception, depression, stressful life events, school functioning and involvement, and adolescent risk behaviors of substance use, sexual behavior, and weapon carrying were measured in the cross-sectional surveys at both schools by means of a selfadministered anonymous questionnaire. Passive pa-
314
KULIG ET AL.
rental consent was implemented for this study protocol, requiring that parents who did not want their child to participate sign a disapproval notice. A predictive model of weapon carrying as the major outcome of interest was developed using a forward stepwise logistic regression technique in School A and tested in School B to determine if the relationship between predictors and outcome was the same in both secondary school populations. Weapon carrying was measured by a dichotomous question which attempted to identify lifetime experience: “Do you ever carry a weapon—yes or no?”
Study Population A total of 659 students responded to the health survey in School A, which represented 68% of the student body enrolled in the school at the time of the survey administration. A total of 1016 students participated in the survey at School B, also representing 68% of the student body. In School A, 29% of the student body was not in attendance on the day of the survey, compared to 27% in School B. However, survey respondents were demographically comparable to absent students at both sites. A comparison of demographic characteristics of absent students and survey respondents at both schools indicated no significant differences in age, race, and gender. Three percent of the surveys in School A and 5% in School B were not usable, as they were incomplete. A negligible number (fewer than 10) of parent refusal forms were returned in each school. As indicated in Table 1, the study populations at both schools were comparable on most demographic measures, with the exceptions of race and gender distribution. On the whole, the students at both schools were from predominantly black and Hispanic background, but School A had a higher proportion of white and Hispanic students and School B had a higher proportion of black students (p , 0.001). School A also had a higher proportion of female students (p , 0.001). There were no other statistically significant differences in other demographic characteristics measured: age, percentage from single parent households, maternal educational attainment, and paternal educational attainment. Statistical significance was determined from Chi-square tests for categorical variables (gender, race, single parent household status, and maternal and paternal education) and student’s t-tests for continuous variables (age).
JOURNAL OF ADOLESCENT HEALTH Vol. 22, No. 4
Table 1. Comparison of demographic characteristics of survey respondents attending two public secondary schools in Boston, 1992 Demographic Characteristics
School A (n 5 659)
School B (n 5 1016)
p Value*
Mean age (SD) Gender distribution Female Male Racial distribution Black Hispanic White Other Living situation Two-parent household Single-parent household Other Maternal education less than high school Paternal education less than high school
16.7 (1.6)
16.9 (1.6)
0.52 ,0.001
57.6% 42.4%
49.5% 50.6%
46.7% 32.8% 5.8% 4.8%
64.3% 22.9% 6.2% 6.6%
45.2%
47.1%
41.8%
39.5%
12.9% 32.7%
13.4% 30.4%
0.40
32.0%
29.9%
0.50
,0.001
0.65
* Results of Student’s t test (for age) or Chi-square test.
Measures A 98-item questionnaire was used to obtain information from participants on self-reported weapon carrying and related measures: demographic profile; psychosocial status; 30-day use of alcohol, tobacco, and other drugs (ATOD); interpersonal violence; sexual behavior; and school involvement. The questionnaire was developed by the Massachusetts Department of Public Health (MDPH) and was used statewide in 13 high schools throughout Massachusetts prior to the 1992 study reported here. Both an English and a Spanish version of the questionnaire were available for participants. The questionnaire was constructed from selected measures in the Centers for Disease Control’s Youth Risk Behavior Survey (5) (to measure alcohol, tobacco, drugs, interpersonal violence), and Maria Kovac’s Children’s Depression Inventory (16) (self-esteem and depression). School involvement measures and the stressful life events list were developed from focus groups of students, parents, and school health specialists conducted by officials of the MDPH and were thus not derived from previously validated instruments. The overall questionnaire was pretested by MDPH officials and survey research consultants with backgrounds in public health, survey research, cultural diversity, and adolescent development. The pretest was carried out on three groups of 10 students
April 1998
randomly selected from schools not included in this study. A test-retest reliability score of 0.81 was obtained in the pretest study. As a result of the pilot test, the questionnaire was translated into two additional languages besides English and Spanish: Haitian Creole and Vietnamese. Demographic variables measured by the survey instrument included: gender, race, grade in school, living arrangement, and parental education. Psychosocial measures included mood (depression) and stressful life events. Measurement of mood included seven items: satisfaction with life, control over life, liking self, feeling sad/upset, number of friends, amount of time spent watching television, and wanting to harm or kill self. Total number of events (total of 16 items) was counted as the Stressful Life Events score (composite scoring from principle component analysis was equally predictive of outcomes, and thus absolute number of events was used as the measure for this analysis). The following health behaviors related to 30-day ATOD use, sexual behavior, and interpersonal violence were measured on the questionnaire and incorporated in this study: a total of 14 items measuring 30-day use of alcohol, tobacco, marijuana and “other” drugs (psychedelics, crack, cocaine, amphetamines, barbiturates, inhalants); 2 items assessing early onset of sexual activity; 1 item each assessing previous pregnancy, hit/beat up someone, witnessed a crime, threatened with a knife/gun, or sustained a serious injury in a fight. School involvement measures included: average grades in the previous semester, skipping school, skipping class, or suspended from school. Statistical Methods Univariate associations between demographic, psychosocial, educational, and behavioral measures and the outcome variable of weapon carrying were tested using the Pearson Chi-square test for discrete parameters and a Wald Chi-square from a logistic regression for continuous parameters. A forward stepwise selection process was used to build a multivariate logistic regression model of factors associated with the weapon-carrying outcome in School A. This selection method was chosen to maximize the sample size in the final model, and hence minimize bias which may be introduced by using only a subset of the full dataset. All parameters with univariate associations with weapon carrying achieving a p value of ,0.05 were considered as candidates for the multivariate analyses with the following two exceptions. Other drug use and sexual intercourse before age 16
WEAPON CARRYING
315
years each had a univariate association with weapon carrying at the p , 0.05 level. However, they were not entered into the multivariate selection process because these measures were missing in more than 10% of the surveys with weapon-carrying information. The potential impact of these two variables is estimated in the model, nonetheless, by including two similar and highly intercorrelated measures for which sufficient data are available: marijuana use and lifetime sexual intercourse, respectively. An adjusted p value of ,0.05 was required for entry into the model at each step of the model-building process. The accuracy of the statistical model to discriminate weapon carriers from non–weapon carriers was quantified using the area under the receive operating characteristic (ROC) curve, represented as the C statistic. If more than one parameter had an adjusted p value of ,0.05, then the parameter with the lowest adjusted p value was selected. If more than one parameter had the same adjusted p value, then the one with the highest C statistic, representing the area under the ROC curve, was entered. The selection process was terminated when no remaining parameters met the criterion for entry into the multivariate model. The predictor variables from the School A model were then used to construct a new multivariate logistic regression model using the School B data. The odds ratios (ORs) for each predictor variable were statistically compared between the School A and School B models with a multivariate logistic regression model using the combined data from both Schools A and B. This model included factors for each of the predictor variables, a factor for the school effect, and all two-factor interaction terms between the predictor variables and the school effect. The interaction terms were used to test the null hypothesis that the odds ratios for each predictive variable were the same between schools. All analyses were performed using the SAS software system for Windows (V6.10).
Results The Predictive Model in School A Among 487 students included in the final multivariate model at School A, 38% of survey respondents reported ever carrying a weapon. Results from the univariate analysis for School A, the population used as the basis for developing the predictive model, identified 31 variables significantly associated with weapon carrying. Significant demographic associations were: male gender, lower grade in school, and
316
KULIG ET AL.
JOURNAL OF ADOLESCENT HEALTH Vol. 22, No. 4
younger age. Notably, race and socioeconomic indicators of maternal and paternal education and family composition were not significantly associated with weapon carrying in the univariate analysis and thus were not candidates for the multivariate model. The sum of stressful life events, wanting to harm self, and “feeling life was not okay” were significantly associated with the outcome variable in the univariate analyses. Significant associations of risk behaviors included: tobacco use, alcohol use, marijuana use, “beat up” someone, threatened with a knife, experienced injury in a fight, history of pregnancy, and sexual experience. School involvement measures significantly associated with the outcome in the univariate analyses included: skipping class, skipping school, suspended from school, and low or failing grades. Of the 31 variables associated with weapon carrying in the univariate analyses, 8 entered into the final main effect model. Seven of these variables were significantly associated with weapon carrying (adjusted p , 0.05). Results of the final predictive model in School A are presented in Table 2. Based upon the multivariate logistic model employed here, predictors of weapon carrying among survey respondents from School A include gender and age, with males significantly more likely than females to report weapon carrying [adjusted OR 5 2.17, 95% confidence interval (CI) 5 1.37, 3.44), and younger students more likely to report weapon carrying than older students (adjusted OR 5 0.81, 95% CI 5 0.69, 0.95). One psychosocial measure remained in the predictive model as significantly associated with weapon carrying. Those students who thought about
Table 2. Results of final predictive models in School A and School B: 1992 survey respondents Odds Ratio† (95% Confidence Interval) Markers in Multivariate Model 30-day marijuana use Sexual experience Witnessed crime Harm self Male gender Beat up someone Skipped school Age (per yr) ROC area
School A (n 5 487)
School B (n 5 702)
p Value*
6.55(3.24, 13.17) 3.38(1.73, 6.63) 2.51(1.42, 4.43) 2.24(1.38, 3.62) 2.17(1.37, 3.44) 1.56(0.94, 2.59) 1.56(1.02, 2.42) 0.81(0.69, 0.95) c 5 0.80
3.20(1.69, 6.05) 3.37(2.00, 5.66) 2.73(1.66, 4.47) 1.50(0.96, 2.23) 1.92(1.29, 2.85) 2.91(1.91, 4.42) 1.52(1.03, 2.25) 0.84(0.74, 0.97) c 5 0.81
0.14 0.99 0.83 0.23 0.69 0.06 0.92 0.70
* Wald Chi-square p value comparing odds ratios between School A and School B. † Adjusted for all other measures in multivariate model.
or wanted to kill themselves were more likely to report weapon carrying than those who did not consider self-harm (adjusted OR 5 2.24, 95% CI 5 1.38, 3.62). Significantly associated risk behaviors included 30-day use of marijuana (adjusted OR 5 6.55, 95% CI 5 3.24, 13.17) being sexually experienced (adjusted OR 5 3.38, 95% CI 5 1.73, 6.63), and having witnessed a crime (adjusted OR 5 2.51, 95% CI 5 1.42, 4.43). The only school involvement measure that remained in the multivariate model was skipping school (adjusted OR 5 1.56, 95% CI 5 1.01, 2.42). Validation of the School A Predictive Factors in School B Self-reported prevalence of weapon carrying among the 702 survey respondents at School B was lower than at School A: 28% versus 38%. A multivariate model was used in School B to determine if the eight predictors identified in the first school would also predict the outcome in another school which is, with the exception of gender and race distribution, demographically comparable to the first school. None of the adjusted ORs from the eight variables used to model weapon carrying among students at School B, as shown in Table 2, were statistically significantly different from the adjusted ORs for the same variables in the model for School A. Demographic characteristics of gender and age continued to predict weapon carrying at the second school. The self-harm variables were marginally significant in School B, with an adjusted OR of 1.5, 95% CI 5 0.96, 2.33. The same risk behaviors that predicted weapon carrying in the first school also predicted weapon carrying in the second school: 30-day marijuana use, sexual experience, and having witnessed a crime. However, the risk behavior of hit/ beat up someone had a marginally significantly (p 5 0.06) greater relationship to weapon carrying in School B compared to School A (adjusted OR 5 2.91, 95% CI 5 1.91, 4.42). The relationship of the school involvement measure of “skip school” in School B (adjusted OR 5 1.52, 95% CI 5 1.03, 2.25) was of the same magnitude as in School A. Even if the different weightings for the relationship between the predictor variables and weapon carrying from the School A and School B logistic regression models are ignored, one can see that in both schools, students with more risk factors were more likely to carry a weapon (Figure 1). This finding again suggests that the variables are independent markers of weapon carrying.
April 1998
Figure 1. Weapon carrying rates stratified by the number of risk factors for School A and School B: 1992 survey respondents.
Discussion The major predictors of weapon carrying among urban secondary school students who are predominantly from black and Hispanic backgrounds are a combination of demographic, psychosocial, behavioral, and school-related factors. This analysis indicates consistency in all eight markers associated with weapon carrying: lower age (a continuous variable ranging from 13–18 years, with a mean age of 16.7 years and 16.9 years in each school respectively), male gender, regular marijuana use, sexual experience, having witnessed a crime, having skipped school, suicidal ideation (significant at the p , 0.05 level in School A only), and having hit or beat up someone (significant at the p , 0.05 in School B only). In both Schools A and B, marijuana use and sexual experience were the variables with the strongest relationships with the outcome of weapon carrying as indicated by the largest adjusted ORs. The model-building and validation presented in this study provide empirical evidence for three important conclusions. First, weapon carrying is associated with multiple and interrelated factors which include demographic, psychosocial, behavioral, and school-related characteristics of high school–age adolescents. Second, students with more risk factors are more likely to carry a weapon, suggesting that the variables are independent markers. Third, this study identified marijuana use and being sexually experienced each to be highly predictive of weapon carrying. Both of these risk behaviors are prevalent in the
WEAPON CARRYING
317
U.S. secondary school population. Approximately one third of U.S. high school students report lifetime marijuana use, and about half report being sexually experienced (5). The multivariate model compares rates of self-reported weapon carrying among those sexually experienced and using marijuana with those who do not report these risk behaviors. Thus, use of a multivariate logistic regression model is useful in determining which risk behaviors, such as marijuana use and sexual experience among high school students, are independent markers of weapon carrying. In addition, ORs can be estimated from the logistic regression model, and these statistics can be used as easily interpretable measures of association between risk factors and the outcome while maintaining the characteristic of being independent of the underlying prevalence of the risk factor in the sample. The findings from this study suggest a possible syndrome or co-occurrence of illegal drug use, school truancy, and interpersonal violence that surrounds weapon carrying, especially for younger students. The association of younger age with weapon carrying found in this study is consistent with findings from other studies reported in the literature. For example, Valois et al. (9) found an OR of ,1 for the association of age with reports of carrying a weapon to school, with statistical significance for white males and females and approaching statistical significance for black males and females. However, further analysis and hypothesis testing are required to better understand the relationship of younger age to reported weapon carrying. The higher overall rate of weapon carrying of 38% among School A students used in the final model is higher than Massachusetts 1995 statewide average of 20% (17) and the 1993 estimated prevalence of 22% nationwide (5) among high school students. The higher rate may reflect the urban location of this study population. In addition, the rate of 38% represents lifetime weapon carrying compared with 30day estimates in these other prevalence studies. The study results are based upon self-report in a selfadministered questionnaire. Possible biases of either underreporting or overreporting can occur with selfreports. However, guaranteed anonymity was used in this study to encourage honest answers. The results of this study must be applied to students in attendance at the schools on the day of the survey administration. It should also be pointed out that the participation rate among students who were in attendance at the school on the survey data was extremely high, as only 3–5% of surveys were not usable, and a negligible number of students were
318
KULIG ET AL.
denied parental permission to participate. Those who were absent were demographically comparable to those in attendance. Thus, the survey-takers are demographically representative of the student body as a whole. Absentee students may be at higher risk than those in attendance, and therefore the results of the study may not be generalizable to secondary students who may have high or frequent absenteeism. It is notable that the factors selected by the regression model as predictive of weapon carrying in School A continued to be predictive of weapon carrying in a sample with a somewhat lower weapon carrying prevalence, namely, School B. The generalizability of the factors selected as predictive of weapon carrying in School A to other schools, even those with lower overall weapon carrying prevalence, is supported by these results. Male gender was a significant factor in weapon carrying, as in all previous studies, but race-specific prevalence rates indicate that weapon carrying is occurring in all racial groups measured in this study. Findings regarding the association between substance abuse and weapon carrying probably vary by geographic area of the study sample. While we found a strong association with marijuana use, others have found no association with substance use (11), or an association with alcohol alone (10). The association of younger age with higher reports of weapon carrying is also consistent with the literature and may be indicative of troubled youth selecting themselves out of high school by 11th or 12th grade. The association of demographic factors of race and age with weapon carrying merit careful methodological attention for future studies. In study populations where several racial groups are represented, as in this study, race was not associated with weapon carrying. Other studies reporting excessively high rates of weapon carrying among racially homogeneous groups (18) should be interpreted cautiously if there is no variation in the underlying racial distribution of the study population. In addition, increased age may be associated with weapon carrying at the junior high level (18), but the reverse seems to be true at the high school level, as suggested in this study. Further exploration of this association is needed. Overall, the findings in this study are consistent with other reported findings and lend further support to a model of co-occurrence of multiple and interrelated risk behaviors among urban, in-school adolescents. The association of psychosocial measures, such as the self-harm measure, suggest the
JOURNAL OF ADOLESCENT HEALTH Vol. 22, No. 4
importance of mental health concerns of high school students. This analysis has identified some important factors associated with weapon carrying that should be further studied for possible causal roles in interpersonal violence. Comprehensive school health programs with multidisciplinary services may hold promise in reducing weapon carrying and associated behaviors, with such services as early identification of students at risk, referral for individual and group mental health services, and implementation of violence prevention and conflict resolution curricula on site. This study contributes information sufficient to test other hypotheses about possible causes of youth violence. Presented at the Annual Meeting of the Society for Adolescent Medicine in Los Angeles, California, March 1994. This research was supported by grants from the Prudential Foundation, Newark, New Jersey, and the Deborah K. Noonan Memorial Fund, Boston, Massachusetts. Cooperation and support of the Massachusetts Department of Public Health, Division of Child and Family Health Services, are acknowledged for instrument development and permission for data analysis.
References 1. Kochanek KD, Hudson BL. Advance Report of Final Mortality Statistics 1992. Monthly Vital Statist Rep 1992;43(6). 2. National Center for Health Statistics. Vital Statistics of the United States 1989, Volume 2: Mortality, Part A. Hyattsville, MD, National Center for Health Statistics, 1992. 3. Centers for Disease Control and Prevention. Deaths resulting from firearm and motor vehicle–related injuries—United States, 1968 –1991. MMWR 1994;43:37. 4. Healthy People 2000: National Health Promotion and Disease Prevention Objectives. Washington, DC, U.S. Department of Health and Human Services—Public Health Service, 1990. 5. Centers for Disease Control and Prevention. Youth Risk Behavior Surveillance—United States, 1993. MMWR 1995; 44(SS1):5. 6. DuRant RH, Cadenhead C, Pendergrast RA, et al. Factors associated with the use of violence among urban black adolescents. Am J Public Health 1994;84:612–7. 7. Webster DW, Gainer PS, Champion HW. Weapon carrying among inner-city junior high school students: Defensive behavior vs. aggressive delinquency. Am J Public Health 1993; 83:1604 – 8. 8. Sosin DM, Koepsell TD, Rivara FP, Mercy JA. Fighting as a marker for multiple problem behaviors in adolescents. J Adolesc Health 1995;16:209 –15. 9. Valois RF, McKeown RE, Garrison CZ, Vincent ML. Correlates of aggressive and violent behaviors among public high school adolescents. J Adolesc Health 1995;16:26 –34. 10. Orpinas PK, Basen-Engquist K, Grunbaum J, Parcel GS. The co-morbidity of violence-related behaviors with health-risk behaviors in a population of high school students. J Adolesc Health 1995;16:216 –25. 11. Callahan CM, Rivara FP. Urban high school youth and handguns: A school-based survey. JAMA 1992;267:3038 – 42.
April 1998
12. DuRant RH, Getts AG, Cadenhead C, Woods ER. The association between weapon carrying and the use of violence among adolescents living in and around public housing. Am J Public Health 1995;17:376 – 80. 13. Jessor R, Chase JA, Donovan JE. Psychosocial correlates of marijuana use and problem drinking in a national sample of adolescents. Am J Public Health 1980;70:604 –13. 14. Irwin CE, Millstein SG. Biopsychosocial correlates of risktaking behaviors during adolescence. J Adolesc Health Care 1986;7:82S–96S. 15. Levitt, MZ, Selman RL, Richmond JB. The psychosocial foun-
WEAPON CARRYING
319
dations of early adolescents’ high-risk behavior: Implications for research and practice. J Res Adolesc 1991;1:349 –78. 16. Kovacs M. The Children’s Depression Inventory. Pharmacol Bull 1985;21:995– 8. 17. 1995 Massachusetts Youth Risk Behavior Survey Results. Commonwealth of Massachusetts, Department of Education, April 1996: 39. 18. Vaughan RD, McCarthy JF, Armstrong B, et al. Carrying and using weapons: A survey of minority junior high school students in New York City. Am J Public Health 1996;86:568 – 72.
ORIGINAL ARTICLE
Predictive Model of Weapon Carrying Among Urban High School Students: Results and Validation JOHN KULIG, M.D., M.P.H., JEANETTE VALENTINE, Ph.D., JOHN GRIFFITH, Ph.D., AND ROBIN RUTHAZER, M.P.H.
Objectives: The purpose of this study was to identify the behavioral, psychosocial, and demographic predictors of self-reported weapon carrying among secondary school students who attend urban public schools. Methods: Self-reported weapon carrying was measured in a schoolwide anonymous health survey conducted in two demographically comparable high schools in 1992, in Boston, Massachusetts. Indicators of self-perception, depression, stressful life events, and adolescent risk behaviors of substance use and sexual behavior, along with self-reported weapon carrying, were measured. The students in both schools were racially heterogeneous, with the majority of about 80% from black or Hispanic backgrounds. A predictive model was developed using a forward stepwise logistic regression model in one innercity high school, and tested in a second high school. Results: Self-reported lifetime weapon carrying was 32% overall. The major predictors of weapon carrying among urban secondary school students are a combination of demographic, psychosocial, behavioral, and school-related factors. This analysis indicates consistency in eight markers predictive of weapon carrying: lower age, male gender, regular marijuana use, sexual experience, having witnessed a crime, having skipped school, suicidal ideation, and having hit or “beat up” someone.
From the Department of Pediatrics, Division of General Pediatrics and Adolescent Medicine (J.K., J.V.), The New England Health and Poverty Action Center (J.V.) Boston, Massachusetts; and the Department of Internal Medicine, Division of Clinical Care Research (J.G., R.R.), New England Medical Center, Boston, Massachusetts Address reprint requests to: John Kulig, M.D., M.P.H., Director of Adolescent Medicine, New England Medical Center, 750 Washington Street, Box 479, Boston, MA 02111. Manuscript accepted July 11, 1997. 1054-139X/98/$19.50 PII S1054(98)00256-5
Race parental education, and family composition were not significant predictors. Significant predictors of weapon carrying were marijuana use and sexual experience, each of which was consistently high in both schools. Conclusions: The model-building and validation presented in this study provide empirical evidence for three important conclusions. First, weapon carrying is associated with multiple and interrelated factors which include demographic, psychosocial, behavioral, and school-related characteristics of high school–age adolescents. Second, students with more risk factors are more likely to carry a weapon, suggesting that the variables are independent markers. Third, this study identified marijuana use and being sexually experienced as both highly predictive of weapon carrying. Implications of this study for prevention point to the need for comprehensive multidisciplinary services in high school that include mental health counseling as well as health education efforts aimed at behavior change. © Society for Adolescent Medicine, 1998 KEY WORDS: Violence Weapon(s) Students High school Survey
Weapon carrying among youth is a significant risk factor associated with adolescent morbidity and mortality. Homicide remains the second leading cause of mortality in the U.S. population ages 15–24,
© Society for Adolescent Medicine, 1998 Published by Elsevier Science Inc., 655 Avenue of the Americas, New York, NY 10010
April 1998
accounting for 8019 deaths in 1992 (1). From 1979 to 1992, the age-specific death rate from homicide rose from 14.2 to 22.2/100,000 population in this age group (1). The rate of 37.7 among males is six times that of the female rate of 6.4, and homicide is the most common cause of death among black males age 15–24, with the rate of 154.4/100,000 population being three times that of deaths owing to motor vehicle and other accidents (49.9) (1). The availability of a weapon in a violent confrontation significantly increases the risk of a lethal outcome. In 1989, 80% of homicide victims aged 10 –19 were killed with guns, and 10% were stabbed to death (2). Based upon present trends, it has been projected that by the year 2003, firearms will surpass motor vehicles as the leading cause of injury-related death in the United States (3). Understanding who carries weapons and why can provide information of value in preventing and reducing the use of weapons among young people. Healthy People 2000 has established a goal of reducing by 20% weapon carrying among adolescents aged 14 –17 (4), estimated to be 22.1% in 1993 from the annual Youth Risk Behavior Surveillance System of the Centers for Disease Control and Prevention (5). Factors associated with weapon carrying have been the subject of numerous investigations aimed at identifying the causes and correlates of this lifethreatening risk behavior. Empirical evidence about the correlates of weapon carrying suggests that carrying weapons is indicative of and associated with violent and aggressive behavior and is used for purposeful criminal activity, rather than self-protection (6,7). Studies analyzing data from the national Youth Risk Behavior Survey find that weapon carrying is intercorrelated with other adolescent risk behaviors, including fighting that involves injuries, substance abuse, school truancy, and sexual activity, thus suggestive of a syndrome of antisocial behavior among troubled youth (5,8 –11). Gender differences in the prevalence of weapon carrying and the reasons for its use are also important considerations. For males, weapon carrying is likely to be associated with violent aggressive behavior in their community, while for females, weapon carrying is likely to be associated with gang fighting (12). A number of studies of weapon carrying among youth are observational cross-sectional ones which rely on correlational or multivariate analysis that is often limited to co-occurring behavioral measures. Information on social and psychological factors is of critical importance to understanding who carries weapons and why. This information can contribute
WEAPON CARRYING
313
to developing preventive interventions for youth at risk of carrying and using weapons. A developmental perspective on adolescent behavior points to the importance of psychological state and social and familial factors that can help explain antisocial and self-harming behaviors of youth (13–15). The purpose of this study was to go beyond correlational or predictive modeling of interrelated behaviors and to identify the association of both behavioral and psychosocial factors with weapon carrying among high school students. The main study question was: What demographic, psychosocial, and behavioral factors are predictive of self-reported weapon carrying among adolescents attending inner-city, urban secondary schools? The study approach develops a predictive model of weapon carrying from two crosssectional high school student surveys that include adolescent risk behaviors identified in the literature as associated with weapon carrying as well as psychosocial and demographic measures.
Methods Study Design Cross-sectional, schoolwide anonymous surveys were conducted in two secondary schools in Boston in the Spring of 1992. The two schools were chosen based upon willingness of school administrators to participate in a schoolwide health survey and demographic comparability. The surveys were carried out by trained survey administrators. The prearranged administration of the survey was accomplished during a class period of 45 min, with several survey administrators going to multiple classrooms simultaneously on the same day. The questionnaires were distributed in each classroom by the survey administrators who responded to questions when needed and collected the completed surveys. Teachers did not participate in survey administration but remained in the classrooms. The survey was a population-based survey with all students in attendance on the day of survey eligible for participation. The size of the student body at each of the two schools differed. The student body at School A was composed of 975 students. The student body at School B was composed of 1500 students. Indicators of selfperception, depression, stressful life events, school functioning and involvement, and adolescent risk behaviors of substance use, sexual behavior, and weapon carrying were measured in the cross-sectional surveys at both schools by means of a selfadministered anonymous questionnaire. Passive pa-
314
KULIG ET AL.
rental consent was implemented for this study protocol, requiring that parents who did not want their child to participate sign a disapproval notice. A predictive model of weapon carrying as the major outcome of interest was developed using a forward stepwise logistic regression technique in School A and tested in School B to determine if the relationship between predictors and outcome was the same in both secondary school populations. Weapon carrying was measured by a dichotomous question which attempted to identify lifetime experience: “Do you ever carry a weapon—yes or no?”
Study Population A total of 659 students responded to the health survey in School A, which represented 68% of the student body enrolled in the school at the time of the survey administration. A total of 1016 students participated in the survey at School B, also representing 68% of the student body. In School A, 29% of the student body was not in attendance on the day of the survey, compared to 27% in School B. However, survey respondents were demographically comparable to absent students at both sites. A comparison of demographic characteristics of absent students and survey respondents at both schools indicated no significant differences in age, race, and gender. Three percent of the surveys in School A and 5% in School B were not usable, as they were incomplete. A negligible number (fewer than 10) of parent refusal forms were returned in each school. As indicated in Table 1, the study populations at both schools were comparable on most demographic measures, with the exceptions of race and gender distribution. On the whole, the students at both schools were from predominantly black and Hispanic background, but School A had a higher proportion of white and Hispanic students and School B had a higher proportion of black students (p , 0.001). School A also had a higher proportion of female students (p , 0.001). There were no other statistically significant differences in other demographic characteristics measured: age, percentage from single parent households, maternal educational attainment, and paternal educational attainment. Statistical significance was determined from Chi-square tests for categorical variables (gender, race, single parent household status, and maternal and paternal education) and student’s t-tests for continuous variables (age).
JOURNAL OF ADOLESCENT HEALTH Vol. 22, No. 4
Table 1. Comparison of demographic characteristics of survey respondents attending two public secondary schools in Boston, 1992 Demographic Characteristics
School A (n 5 659)
School B (n 5 1016)
p Value*
Mean age (SD) Gender distribution Female Male Racial distribution Black Hispanic White Other Living situation Two-parent household Single-parent household Other Maternal education less than high school Paternal education less than high school
16.7 (1.6)
16.9 (1.6)
0.52 ,0.001
57.6% 42.4%
49.5% 50.6%
46.7% 32.8% 5.8% 4.8%
64.3% 22.9% 6.2% 6.6%
45.2%
47.1%
41.8%
39.5%
12.9% 32.7%
13.4% 30.4%
0.40
32.0%
29.9%
0.50
,0.001
0.65
* Results of Student’s t test (for age) or Chi-square test.
Measures A 98-item questionnaire was used to obtain information from participants on self-reported weapon carrying and related measures: demographic profile; psychosocial status; 30-day use of alcohol, tobacco, and other drugs (ATOD); interpersonal violence; sexual behavior; and school involvement. The questionnaire was developed by the Massachusetts Department of Public Health (MDPH) and was used statewide in 13 high schools throughout Massachusetts prior to the 1992 study reported here. Both an English and a Spanish version of the questionnaire were available for participants. The questionnaire was constructed from selected measures in the Centers for Disease Control’s Youth Risk Behavior Survey (5) (to measure alcohol, tobacco, drugs, interpersonal violence), and Maria Kovac’s Children’s Depression Inventory (16) (self-esteem and depression). School involvement measures and the stressful life events list were developed from focus groups of students, parents, and school health specialists conducted by officials of the MDPH and were thus not derived from previously validated instruments. The overall questionnaire was pretested by MDPH officials and survey research consultants with backgrounds in public health, survey research, cultural diversity, and adolescent development. The pretest was carried out on three groups of 10 students
April 1998
randomly selected from schools not included in this study. A test-retest reliability score of 0.81 was obtained in the pretest study. As a result of the pilot test, the questionnaire was translated into two additional languages besides English and Spanish: Haitian Creole and Vietnamese. Demographic variables measured by the survey instrument included: gender, race, grade in school, living arrangement, and parental education. Psychosocial measures included mood (depression) and stressful life events. Measurement of mood included seven items: satisfaction with life, control over life, liking self, feeling sad/upset, number of friends, amount of time spent watching television, and wanting to harm or kill self. Total number of events (total of 16 items) was counted as the Stressful Life Events score (composite scoring from principle component analysis was equally predictive of outcomes, and thus absolute number of events was used as the measure for this analysis). The following health behaviors related to 30-day ATOD use, sexual behavior, and interpersonal violence were measured on the questionnaire and incorporated in this study: a total of 14 items measuring 30-day use of alcohol, tobacco, marijuana and “other” drugs (psychedelics, crack, cocaine, amphetamines, barbiturates, inhalants); 2 items assessing early onset of sexual activity; 1 item each assessing previous pregnancy, hit/beat up someone, witnessed a crime, threatened with a knife/gun, or sustained a serious injury in a fight. School involvement measures included: average grades in the previous semester, skipping school, skipping class, or suspended from school. Statistical Methods Univariate associations between demographic, psychosocial, educational, and behavioral measures and the outcome variable of weapon carrying were tested using the Pearson Chi-square test for discrete parameters and a Wald Chi-square from a logistic regression for continuous parameters. A forward stepwise selection process was used to build a multivariate logistic regression model of factors associated with the weapon-carrying outcome in School A. This selection method was chosen to maximize the sample size in the final model, and hence minimize bias which may be introduced by using only a subset of the full dataset. All parameters with univariate associations with weapon carrying achieving a p value of ,0.05 were considered as candidates for the multivariate analyses with the following two exceptions. Other drug use and sexual intercourse before age 16
WEAPON CARRYING
315
years each had a univariate association with weapon carrying at the p , 0.05 level. However, they were not entered into the multivariate selection process because these measures were missing in more than 10% of the surveys with weapon-carrying information. The potential impact of these two variables is estimated in the model, nonetheless, by including two similar and highly intercorrelated measures for which sufficient data are available: marijuana use and lifetime sexual intercourse, respectively. An adjusted p value of ,0.05 was required for entry into the model at each step of the model-building process. The accuracy of the statistical model to discriminate weapon carriers from non–weapon carriers was quantified using the area under the receive operating characteristic (ROC) curve, represented as the C statistic. If more than one parameter had an adjusted p value of ,0.05, then the parameter with the lowest adjusted p value was selected. If more than one parameter had the same adjusted p value, then the one with the highest C statistic, representing the area under the ROC curve, was entered. The selection process was terminated when no remaining parameters met the criterion for entry into the multivariate model. The predictor variables from the School A model were then used to construct a new multivariate logistic regression model using the School B data. The odds ratios (ORs) for each predictor variable were statistically compared between the School A and School B models with a multivariate logistic regression model using the combined data from both Schools A and B. This model included factors for each of the predictor variables, a factor for the school effect, and all two-factor interaction terms between the predictor variables and the school effect. The interaction terms were used to test the null hypothesis that the odds ratios for each predictive variable were the same between schools. All analyses were performed using the SAS software system for Windows (V6.10).
Results The Predictive Model in School A Among 487 students included in the final multivariate model at School A, 38% of survey respondents reported ever carrying a weapon. Results from the univariate analysis for School A, the population used as the basis for developing the predictive model, identified 31 variables significantly associated with weapon carrying. Significant demographic associations were: male gender, lower grade in school, and
316
KULIG ET AL.
JOURNAL OF ADOLESCENT HEALTH Vol. 22, No. 4
younger age. Notably, race and socioeconomic indicators of maternal and paternal education and family composition were not significantly associated with weapon carrying in the univariate analysis and thus were not candidates for the multivariate model. The sum of stressful life events, wanting to harm self, and “feeling life was not okay” were significantly associated with the outcome variable in the univariate analyses. Significant associations of risk behaviors included: tobacco use, alcohol use, marijuana use, “beat up” someone, threatened with a knife, experienced injury in a fight, history of pregnancy, and sexual experience. School involvement measures significantly associated with the outcome in the univariate analyses included: skipping class, skipping school, suspended from school, and low or failing grades. Of the 31 variables associated with weapon carrying in the univariate analyses, 8 entered into the final main effect model. Seven of these variables were significantly associated with weapon carrying (adjusted p , 0.05). Results of the final predictive model in School A are presented in Table 2. Based upon the multivariate logistic model employed here, predictors of weapon carrying among survey respondents from School A include gender and age, with males significantly more likely than females to report weapon carrying [adjusted OR 5 2.17, 95% confidence interval (CI) 5 1.37, 3.44), and younger students more likely to report weapon carrying than older students (adjusted OR 5 0.81, 95% CI 5 0.69, 0.95). One psychosocial measure remained in the predictive model as significantly associated with weapon carrying. Those students who thought about
Table 2. Results of final predictive models in School A and School B: 1992 survey respondents Odds Ratio† (95% Confidence Interval) Markers in Multivariate Model 30-day marijuana use Sexual experience Witnessed crime Harm self Male gender Beat up someone Skipped school Age (per yr) ROC area
School A (n 5 487)
School B (n 5 702)
p Value*
6.55(3.24, 13.17) 3.38(1.73, 6.63) 2.51(1.42, 4.43) 2.24(1.38, 3.62) 2.17(1.37, 3.44) 1.56(0.94, 2.59) 1.56(1.02, 2.42) 0.81(0.69, 0.95) c 5 0.80
3.20(1.69, 6.05) 3.37(2.00, 5.66) 2.73(1.66, 4.47) 1.50(0.96, 2.23) 1.92(1.29, 2.85) 2.91(1.91, 4.42) 1.52(1.03, 2.25) 0.84(0.74, 0.97) c 5 0.81
0.14 0.99 0.83 0.23 0.69 0.06 0.92 0.70
* Wald Chi-square p value comparing odds ratios between School A and School B. † Adjusted for all other measures in multivariate model.
or wanted to kill themselves were more likely to report weapon carrying than those who did not consider self-harm (adjusted OR 5 2.24, 95% CI 5 1.38, 3.62). Significantly associated risk behaviors included 30-day use of marijuana (adjusted OR 5 6.55, 95% CI 5 3.24, 13.17) being sexually experienced (adjusted OR 5 3.38, 95% CI 5 1.73, 6.63), and having witnessed a crime (adjusted OR 5 2.51, 95% CI 5 1.42, 4.43). The only school involvement measure that remained in the multivariate model was skipping school (adjusted OR 5 1.56, 95% CI 5 1.01, 2.42). Validation of the School A Predictive Factors in School B Self-reported prevalence of weapon carrying among the 702 survey respondents at School B was lower than at School A: 28% versus 38%. A multivariate model was used in School B to determine if the eight predictors identified in the first school would also predict the outcome in another school which is, with the exception of gender and race distribution, demographically comparable to the first school. None of the adjusted ORs from the eight variables used to model weapon carrying among students at School B, as shown in Table 2, were statistically significantly different from the adjusted ORs for the same variables in the model for School A. Demographic characteristics of gender and age continued to predict weapon carrying at the second school. The self-harm variables were marginally significant in School B, with an adjusted OR of 1.5, 95% CI 5 0.96, 2.33. The same risk behaviors that predicted weapon carrying in the first school also predicted weapon carrying in the second school: 30-day marijuana use, sexual experience, and having witnessed a crime. However, the risk behavior of hit/ beat up someone had a marginally significantly (p 5 0.06) greater relationship to weapon carrying in School B compared to School A (adjusted OR 5 2.91, 95% CI 5 1.91, 4.42). The relationship of the school involvement measure of “skip school” in School B (adjusted OR 5 1.52, 95% CI 5 1.03, 2.25) was of the same magnitude as in School A. Even if the different weightings for the relationship between the predictor variables and weapon carrying from the School A and School B logistic regression models are ignored, one can see that in both schools, students with more risk factors were more likely to carry a weapon (Figure 1). This finding again suggests that the variables are independent markers of weapon carrying.
April 1998
Figure 1. Weapon carrying rates stratified by the number of risk factors for School A and School B: 1992 survey respondents.
Discussion The major predictors of weapon carrying among urban secondary school students who are predominantly from black and Hispanic backgrounds are a combination of demographic, psychosocial, behavioral, and school-related factors. This analysis indicates consistency in all eight markers associated with weapon carrying: lower age (a continuous variable ranging from 13–18 years, with a mean age of 16.7 years and 16.9 years in each school respectively), male gender, regular marijuana use, sexual experience, having witnessed a crime, having skipped school, suicidal ideation (significant at the p , 0.05 level in School A only), and having hit or beat up someone (significant at the p , 0.05 in School B only). In both Schools A and B, marijuana use and sexual experience were the variables with the strongest relationships with the outcome of weapon carrying as indicated by the largest adjusted ORs. The model-building and validation presented in this study provide empirical evidence for three important conclusions. First, weapon carrying is associated with multiple and interrelated factors which include demographic, psychosocial, behavioral, and school-related characteristics of high school–age adolescents. Second, students with more risk factors are more likely to carry a weapon, suggesting that the variables are independent markers. Third, this study identified marijuana use and being sexually experienced each to be highly predictive of weapon carrying. Both of these risk behaviors are prevalent in the
WEAPON CARRYING
317
U.S. secondary school population. Approximately one third of U.S. high school students report lifetime marijuana use, and about half report being sexually experienced (5). The multivariate model compares rates of self-reported weapon carrying among those sexually experienced and using marijuana with those who do not report these risk behaviors. Thus, use of a multivariate logistic regression model is useful in determining which risk behaviors, such as marijuana use and sexual experience among high school students, are independent markers of weapon carrying. In addition, ORs can be estimated from the logistic regression model, and these statistics can be used as easily interpretable measures of association between risk factors and the outcome while maintaining the characteristic of being independent of the underlying prevalence of the risk factor in the sample. The findings from this study suggest a possible syndrome or co-occurrence of illegal drug use, school truancy, and interpersonal violence that surrounds weapon carrying, especially for younger students. The association of younger age with weapon carrying found in this study is consistent with findings from other studies reported in the literature. For example, Valois et al. (9) found an OR of ,1 for the association of age with reports of carrying a weapon to school, with statistical significance for white males and females and approaching statistical significance for black males and females. However, further analysis and hypothesis testing are required to better understand the relationship of younger age to reported weapon carrying. The higher overall rate of weapon carrying of 38% among School A students used in the final model is higher than Massachusetts 1995 statewide average of 20% (17) and the 1993 estimated prevalence of 22% nationwide (5) among high school students. The higher rate may reflect the urban location of this study population. In addition, the rate of 38% represents lifetime weapon carrying compared with 30day estimates in these other prevalence studies. The study results are based upon self-report in a selfadministered questionnaire. Possible biases of either underreporting or overreporting can occur with selfreports. However, guaranteed anonymity was used in this study to encourage honest answers. The results of this study must be applied to students in attendance at the schools on the day of the survey administration. It should also be pointed out that the participation rate among students who were in attendance at the school on the survey data was extremely high, as only 3–5% of surveys were not usable, and a negligible number of students were
318
KULIG ET AL.
denied parental permission to participate. Those who were absent were demographically comparable to those in attendance. Thus, the survey-takers are demographically representative of the student body as a whole. Absentee students may be at higher risk than those in attendance, and therefore the results of the study may not be generalizable to secondary students who may have high or frequent absenteeism. It is notable that the factors selected by the regression model as predictive of weapon carrying in School A continued to be predictive of weapon carrying in a sample with a somewhat lower weapon carrying prevalence, namely, School B. The generalizability of the factors selected as predictive of weapon carrying in School A to other schools, even those with lower overall weapon carrying prevalence, is supported by these results. Male gender was a significant factor in weapon carrying, as in all previous studies, but race-specific prevalence rates indicate that weapon carrying is occurring in all racial groups measured in this study. Findings regarding the association between substance abuse and weapon carrying probably vary by geographic area of the study sample. While we found a strong association with marijuana use, others have found no association with substance use (11), or an association with alcohol alone (10). The association of younger age with higher reports of weapon carrying is also consistent with the literature and may be indicative of troubled youth selecting themselves out of high school by 11th or 12th grade. The association of demographic factors of race and age with weapon carrying merit careful methodological attention for future studies. In study populations where several racial groups are represented, as in this study, race was not associated with weapon carrying. Other studies reporting excessively high rates of weapon carrying among racially homogeneous groups (18) should be interpreted cautiously if there is no variation in the underlying racial distribution of the study population. In addition, increased age may be associated with weapon carrying at the junior high level (18), but the reverse seems to be true at the high school level, as suggested in this study. Further exploration of this association is needed. Overall, the findings in this study are consistent with other reported findings and lend further support to a model of co-occurrence of multiple and interrelated risk behaviors among urban, in-school adolescents. The association of psychosocial measures, such as the self-harm measure, suggest the
JOURNAL OF ADOLESCENT HEALTH Vol. 22, No. 4
importance of mental health concerns of high school students. This analysis has identified some important factors associated with weapon carrying that should be further studied for possible causal roles in interpersonal violence. Comprehensive school health programs with multidisciplinary services may hold promise in reducing weapon carrying and associated behaviors, with such services as early identification of students at risk, referral for individual and group mental health services, and implementation of violence prevention and conflict resolution curricula on site. This study contributes information sufficient to test other hypotheses about possible causes of youth violence. Presented at the Annual Meeting of the Society for Adolescent Medicine in Los Angeles, California, March 1994. This research was supported by grants from the Prudential Foundation, Newark, New Jersey, and the Deborah K. Noonan Memorial Fund, Boston, Massachusetts. Cooperation and support of the Massachusetts Department of Public Health, Division of Child and Family Health Services, are acknowledged for instrument development and permission for data analysis.
References 1. Kochanek KD, Hudson BL. Advance Report of Final Mortality Statistics 1992. Monthly Vital Statist Rep 1992;43(6). 2. National Center for Health Statistics. Vital Statistics of the United States 1989, Volume 2: Mortality, Part A. Hyattsville, MD, National Center for Health Statistics, 1992. 3. Centers for Disease Control and Prevention. Deaths resulting from firearm and motor vehicle–related injuries—United States, 1968 –1991. MMWR 1994;43:37. 4. Healthy People 2000: National Health Promotion and Disease Prevention Objectives. Washington, DC, U.S. Department of Health and Human Services—Public Health Service, 1990. 5. Centers for Disease Control and Prevention. Youth Risk Behavior Surveillance—United States, 1993. MMWR 1995; 44(SS1):5. 6. DuRant RH, Cadenhead C, Pendergrast RA, et al. Factors associated with the use of violence among urban black adolescents. Am J Public Health 1994;84:612–7. 7. Webster DW, Gainer PS, Champion HW. Weapon carrying among inner-city junior high school students: Defensive behavior vs. aggressive delinquency. Am J Public Health 1993; 83:1604 – 8. 8. Sosin DM, Koepsell TD, Rivara FP, Mercy JA. Fighting as a marker for multiple problem behaviors in adolescents. J Adolesc Health 1995;16:209 –15. 9. Valois RF, McKeown RE, Garrison CZ, Vincent ML. Correlates of aggressive and violent behaviors among public high school adolescents. J Adolesc Health 1995;16:26 –34. 10. Orpinas PK, Basen-Engquist K, Grunbaum J, Parcel GS. The co-morbidity of violence-related behaviors with health-risk behaviors in a population of high school students. J Adolesc Health 1995;16:216 –25. 11. Callahan CM, Rivara FP. Urban high school youth and handguns: A school-based survey. JAMA 1992;267:3038 – 42.
April 1998
12. DuRant RH, Getts AG, Cadenhead C, Woods ER. The association between weapon carrying and the use of violence among adolescents living in and around public housing. Am J Public Health 1995;17:376 – 80. 13. Jessor R, Chase JA, Donovan JE. Psychosocial correlates of marijuana use and problem drinking in a national sample of adolescents. Am J Public Health 1980;70:604 –13. 14. Irwin CE, Millstein SG. Biopsychosocial correlates of risktaking behaviors during adolescence. J Adolesc Health Care 1986;7:82S–96S. 15. Levitt, MZ, Selman RL, Richmond JB. The psychosocial foun-
WEAPON CARRYING
319
dations of early adolescents’ high-risk behavior: Implications for research and practice. J Res Adolesc 1991;1:349 –78. 16. Kovacs M. The Children’s Depression Inventory. Pharmacol Bull 1985;21:995– 8. 17. 1995 Massachusetts Youth Risk Behavior Survey Results. Commonwealth of Massachusetts, Department of Education, April 1996: 39. 18. Vaughan RD, McCarthy JF, Armstrong B, et al. Carrying and using weapons: A survey of minority junior high school students in New York City. Am J Public Health 1996;86:568 – 72.