Presentation on theme: "Childhood exposure to domestic violence predicts relationship violence: A meta-analysis Markus Kemmelmeier and Kerry Kleyman."— Presentation transcript:
Childhood exposure to domestic violence predicts relationship violence: A meta-analysis Markus Kemmelmeier and Kerry Kleyman
Introduction Problem of violence and abuse in families –Immediate victims –Witnesses (not immediately victimized) –Short-term vs. long-term effects Short-term effects victims Long-term effects victims Short-term/concurrent effects witnesses (e.g., Kitzmann et al., 2003) What about long-term effects in witnesses?
The Cycle of Violence Hypothesis Transmission of violence across generations –Parents/caregivers are an important influence Child Witness Adult Perpetrator Child Witness Adult Victim
Causal processes Social Learning Theory –Relationship norms –Gender norms –Norms for violent behavior –Relationship models Psychodynamic approaches
The “comorbidity” problem Child is not only witness, but also –Victim –in dysfunctional family with –sexual abuse –alcohol abuse –drug abuse
Research field Scattered –Psychology –Social work –Sociology –Medicine –Criminal Justice –Etc. Difference in focus, methods, findings Needed: An integration & review
Meta-analysis Statistical synthesis of available research findings Establishes cumulative science, –Can findings be generalized? –Moderators? Hypothesis testing in the aggregate But: “Garbage in, garbage out” –A meta-analysis is only as good as the data on which it is based
The “How to” of a meta-analysis Define problem, concepts based on literature Decide on inclusion (exclusion) criteria –Compare apples with apples Find studies –Databases –Asking colleagues, listservs –Ancestry approach Code statistics –Effect sizes –Study characteristics File drawer problem
The “How to” of a meta-analysis Statistical size of the effect that is independent of sample size and specific measure used Types r family (Pearson correlation) d family (Pearson correlation): Mean difference divided by standard deviation Extract effect sizes –“read off” –compute –Transform –Infer Research reports might be --incomplete --wrong --unusable
r -- Binomial Effect Size Display (BESD) Transform variables into dichotomous categories Correlation between being a witness and engaging in relationship violence r =.20 –Only 4% of the variance (r 2 ) Risk witnesses 60% Risk non-witnesses 40% Witnesses are 50% more likely to become violent than nonwitnesses.
Method: Literature Search PsychINFO Database Keywords: [(dating or courtship) AND (violence or abuse or aggression)] 1008 identified, 283 included family-of-origin violence Additional studies obtained from references and manual inspection of violence journals
Method: Inclusion/Exclusion Criteria 1.Contained witnessing or experiencing parental violence and perpetrated or experienced violence in adult relationships 2.Included data on physical violence in family of origin or in current dating experience 3.Had to report the quantitative data necessary to calculate at least one effect size 4.Studies had to be reported in English between 1975 and 2006 53 research reports were retained for coding
Method: Coding Article-level coding, including: –Author gender, department affiliation, study design, location, sampling, year of publication Construct-level coding, including: –Sample characteristics, theoretical constructs, methodological variations, and effect size Each article was coded with on article-level, and most had multiple construct-level coding. Two independent coders were used
Method: Data Analysis 402 effect-size estimates from 53 studies Effect-size estimates (r coefficients) were calculated from a variety of reported statistics, including means and standard deviations, chi-square values, p-values, and frequencies or proportions. (Used program Dstat) Correlational studies, point-biserial correlations or Pearson’s r were recorded as individual effect sizes. Because normal distributions of coefficients could often not be assumed, we used a randomization/resampling (bootstrapping) approach to estimate statistical parameters. (Using program MetaWin)
Check for File drawer problem Across all studies coded, is the distribution of the effect sizes –Symmetrical –Funnel shaped?
Witnessing only vs. Experiencing only Witnessing violence has as negative effects as experiencing physical violence. Additive effect? Type of Violence –Q b (3) = 23.17 *** /+ Types of Childhood Violence: Witnessing vs. Experiencing Effect Sizes + Note: +p<.10 * p<.05 ** p<.01 *** p<.001
Results: Overall Witnessing Overall Effects (weighted means) Victimization r =.107 (CI.079/.137) [74 data points] Perpetration r =.138 (CI.113/.162) [116 data points] Unspecified r =.112 (CI.030/.212) [7 data points] Q b (2) = 8.02, p <.28. Witnessing family violence is as strongly linked to becoming a victim of relationship violence as it is to inflicting violence onto others.
Witnessing by Gender Witnessing had stronger effects on men becoming perpetrators No gender difference for victimization. Victimization Q b (1) = 0.59 ns/ns Perpetration Q b (1) = 23.85 ***/** Victimization vs. Perpetration by Gender Effect Sizes ns ** Note: +p<.10 * p<.05 ** p<.01 *** p<.001
Perpetrator in Family of Origin by Gender, VICTIM Witnessing the mother perpetrate violence had the strongest effect on becoming a victim of later relationship violence. Witnessing father violence had a weaker effect on men becoming victims. Male/Father –Q b (1) = 2.39 ns/ns Female/Mother –Q b (1) = 0.65 ns/ns Unspecified –Q b (1) = 0.19 ns/ns Victimization: Perpetrator in Family of Origin by Gender Effect Sizes ns Note: +p<.10 * p<.05 ** p<.01 *** p<.001
Perpetrator in Family of Origin by Gender, PERP Whether male or female witnesses become perpetrators does not depend on whom they witnessed in their family of origin Male/Father –Q b (1) = 3.30 +/ns Female/Mother –Q b (1) = 0.49 ns/ns Unspecified –Q b (1) = 21.32 ***/** Perpetration: Perpetrator in Family of Origin by Gender Effect Sizes ns Note: +p<.10 * p<.05 ** p<.01 *** p<.001
Witnessing by SES Witnessing had the strongest effect on becoming a perpetrator in high SES samples SES had no influence on whether witnesses became themselves victims of relationship violence Victimization Q b (3) = 1.19 ns/ns Perpetration Q b (4) = 29.81 *** /+ Victimization vs. Perpetration by SES Effect Sizes ns + Not enough “high” cases for analysis Note: +p<.10 * p<.05 ** p<.01 *** p<.001
Departmental Affiliation of 1 st Author The background and affiliation of a study’s first author produce great differences in the effect sizes obtained. This suggests that research training and goals have a substantial influence on outcomes. Departmental Affiliation –Q b (4) = 48.30 ***/** Departmental Affiliation Effect Sizes ** Note: +p<.10 * p<.05 ** p<.01 *** p<.001
Physical vs. Psychological Violence Witnessing violence in one’s family of origin make witnesses more likely to perpetrate psychological violence in their own relationships. Victimization –Qb(1) = 0.01 ns/ns Perpetration –Qb(1) = 16.42 ***/* Type of Violence: Physical vs. Psychological Violence Effect Sizes ns * Note: +p<.10 * p<.05 ** p<.01 *** p<.001
Meta-analysis of coefficients comparing region and exposure to family violence to non-exposure Region Q b (5) = 91.27 ***/*** Mean WeightedMean effect size (ri+) RegionkN+-=95% CIHomogeneit y Northeast4284563273.119.094/.14251.518 South51107014560.119.093/.147105.578*** Midwest51113304083.083.055/.113126.512*** West10915712.199.093/.29721.183* Multi-State2822200.305.300/.310.025 Non-U.S.41499328130.209.163/.24480.158*** Note: +p<.10 * p<.05 ** p<.01 *** p<.001
Discussion Witnessing has pervasive effects on likelihood of becoming –a perpetrator –a victim Of relationship violence Long-term effects! Gender effects limited –Mainly males becoming perpetrators
Discussion Other moderators –Class –Discipline Social Learning Theory can help explain –BUT very limited support for a gender specificity hypothesis Limitations