Examining the Reciprocal Nature of the Health-Violence Relationship: Results from a Nationally Representative Sample

Abstract

The correlation between health and offending is typically regarded as the result of confounding factors such as socioeconomic status and drug use, with little consideration given to the plausibility of reciprocal effects. Using two waves of data collected on 14,738 adolescents from the National Longitudinal Study of Adolescent Health (Add Health), a simultaneous structural equations modeling approach was used to determine whether there is a symbiotic relationship between health and violence. Findings indicate that minor health problems have delayed effects on violence and that involvement in violence also negatively affects future health. Discussion centers on this reciprocal relationship, implications for future research, and public health and delinquency prevention policy.

Keywords:

• health
• illness
• violence
• delinquency
• adolescence

Notes

1. We also note that while health problems are likely to lead to stress or strain, they also may result from it. Stress has been linked to increased somatic problems in both children and adolescents (Evans & English, 2002; Haugland, Wold, & Torsheim, 2003; Shephard & Kashani, 1991). Minor somatic and visceral complaints may also be symptoms of underlying problems. Those who visit pediatricians for somatic complaints with no underlying organic etiology have been shown to have experienced significantly more negative life changes than other patient classifications (routine checkups, diagnosed illness, etc.) (Greene, Walker, Hickson, & Thompson, 1985).

2. Jang’s (2007) study utilized a sample of adult African-Americans with a mean age of 43. Therefore, the majority of those within the sample were beyond the ages associated with the majority of offending. Since many of the health strain measures utilized within that study typically only affect adults (heart attack, hardening of the arteries, ulcers, stroke, etc.), it would be inappropriate to use those measures in an adolescent population such as the one used in the present analysis.

3. During the first wave, information was collected from individuals in grades 7-12. In the second wave, data were collected from those same individuals that were in a grade between 7 and 11 at the time of the first wave. Each of the students who graduated in the interval between the two waves was therefore dropped. We do not consider these cases missing, but that the sample is of students grades 7-11 at the time of Wave I. Due to concerns that individuals may have dropped out of the study for other reasons, we compared sample characteristics of our sample to the Wave I sample and found few discrepancies. The total sample had a higher mean age than the sample utilized (as would be expected due to the fact 12th graders are a large portion of the missing) and those dropped had a higher mean income than those in the final sample, but the samples were similar in terms of race and gender composition. In regard to missing data on specific items, we used MPLUS’s default of using all available data to estimate models. The cases with missing weights were excluded from the models presented in the manuscript as MPLUS cannot estimate models in which some cases have missing weights. We reestimated these models with the entire sample, not correcting for weights so as to include these cases and found substantively equivalent results.

4. Nonviolent offenses were also examined using 11 additional delinquency items as indicators (painting graffiti, shoplifting, petty theft, serious theft, destruction of property, burglary, etc.). These items were measured using an identical ordinal scale. Factor analysis revealed delinquency as a second-order factor indicated by two first-order factors (violence and nonviolent delinquency). As our interest lies in violence and the impact of health problems, we do not present models within the text that include these items and instead used this information to confirm that the correct constellation of items was selected to represent violence. Separate models (not presented) failed to show a reciprocal relationship between health and property offenses.

5. Models were also estimated without the correlated errors for repeated measures terms. These fit less adequately than those presented in the text, but the pathways between health and delinquency in each were identical to those in the presented models in terms of sign and significance. The only difference between the two sets of models was that the pathway between Wave I violence and Wave II substance use failed to reach significance in Model 3 without the correlated errors for repeated measures.

6. All effects listed in the text and figures are standardized. Unstandardized estimates are available in the tables.

7. For the sake of clarity, all the additional factors are placed in one box. Note that each of these has independent pathways to the variables of interest and some were measured as latent constructs with multiple indicators (socioeconomic status) while others were appropriately treated as observed variables (age, gender, race, etc.). Those controls that logic dictated should be allowed to correlate with one another (i.e. race and SES) were and those that should be unrelated (i.e. age and gender) were not.

8. Health problems were also associated with conditioning factors that would make deviance more likely such as associating with deviant peers (r = .185, p = .000), low constraint (r = .099, p = .000), low social support (r = −.248, p = .000), and a lack of religiosity (r = −.045, p = .000).

9. The Add Health data-set does contain measures that have been used to represent deviant peers. This three item scale, however, only assesses peer use of alcohol, tobacco, and marijuana, which would be linked to the individuals’ own use and—due to the effect of these substances on health—their health.

Additional information

Notes on contributors

John Stogner

John Stogner is an assistant professor of criminology at Georgia Southern University. His research interests include the relationship between health and delinquency, drug use and drug policy, general strain theory, biosocial theories, and quantitative methodology. His works have appeared in the Journal of Research in Crime and Delinquency, Substance Use and Misuse, Deviant Behavior, the Journal of Youth and Adolescence, and the Journal of Criminal Justice.

Chris L. Gibson

Chris L. Gibson is an associate professor of criminology in the Department of Sociology and Criminology & Law at the University of Florida and a W.E.B. Du Bois Fellow at the National Institute of Justice. His research focuses on the independent and interactive influences of traits and environmental contexts on antisocial behaviors, victimization, and the societal consequences of crime.

J. Mitchell Miller

J. Mitchell Miller is a professor in the Department of Criminal Justice at the University of Texas, San Antonio. He received his PhD in sociology from the University of Tennessee in 1996. He teaches and researches in the areas of drugs and crime, juvenile delinquency, and criminological theory.