CHAPTER FIVE (part 5)

Table 27. Pearson Correlation Coefficients: Personality measures and perpetrated partner abuse comparing Wave 1 and Wave 2 data based on the same sample of female respondents

Note: * p < .05, ** p < .01, *** p < .001

+ Z Scores were derived from the prevalence of partner abuse data and personality measures from Wave 1 and Wave 2, respectively.

Asymptotic Standard Errors are presented in brackets.

Partner abuse was based on the full measure prior to any transformations being conducted.

Testing the Diathesis Stress Model

Logistic Regression Analyses

The rationale behind the application of logistic regression analyses in the testing of the diathesis-stress model of partner abuse has already been discussed.  In summary, the use of this approach provided the opportunity to test the diathesis-stress model of partner abuse by investigating the relative effects of underlying and situational factors on current partner abuse.  Through this  statistical approach, the additive effects of the independent measures as well as the overall fit of the models tested were  assessed.

Several preparatory measures were undertaken prior to conducting  the logistic regressions.  First, the dependent measure, "current perpetrated partner abuse" which assesses the number of times a  respondent perpetrated partner abuse during the past year was collapsed into two levels, "abuse" (1) and "no abuse" (2). Second, the dummy codings previously applied to the demographic variables (i.e., race, religion, employment status) were likewise applied to these analyses.  The transformations previously conducted on the variable "education status" were also utilized in these logistic regression analyses.  Finally, the variables included in these  analyses were inspected for missing data.  Two cases in the male data  and eleven cases in the female data were deleted where systematic missing data were found across a number of variables.  Where missing  data were found on one or two variables only, variables means were  substituted when appropriate.

Analyses were first conducted on the main effects of the diathesis-stress model of partner abuse for both males and females.
     The diathesis measures included demographic variables (Wave 1),  personality measures (Wave 1), past perpetrated partner abuse (prevalence of perpetrated partner abuse reported in Wave 1),  violence in the family of origin (Wave 2) and the alcohol dependence  index (Wave 1).  The stress measures included the life events stress  scale (Wave 2) and current alcohol consumption (Wave 2).  Two  separate interaction models were also tested.  One assessed the  diathesis by life stress events effects, and the other assessed the  diathesis by alcohol consumption effects.

Tables 29 and 30 provide the results of the logistic regressions  conducted on the variables included in the diathesis-stress models  for males and females, respectively.  The results are organized as follows.  The parameter estimates for the main effects  diathesis-stress model of partner abuse are presented in Column 1.
     The parameter estimates for the interaction models are presented in Column 2 (life stress events interactions) and in Column 3 (alcohol interactions).  Only variables found to be significant in the main  effects models were included in the interaction terms.  The last column presents the antilogs for the main effects listed in Column 1  (i.e., relative odds ratios).

Male data

The results of the logistic analyses conducted on the  male data are presented in Table 29.  The coefficients in Column 1  indicate that of the 19 diathesis and stress variables tested, the  following five had statistically significant effects on the log odds  of currently perpetrated partner abuse by males: age, Catholic, past  perpetrated partner abuse, observing father hitting mother, and  stress.  The parameter estimates indicated that males who are young  in age, non-Catholic, have perpetrated past partner abuse, have observed their fathers hitting their mothers, and who experience high levels of stress are more likely to perpetrate current partner abuse than those not identified by these factors.  The global test for the  significance of the diathesis-stress main effects model was L2=69.056  with 19 d.f., indicating that it is highly significant (p < .0001).

The interaction models in Columns 2 and 3 addressed the question, "does stress and alcohol consumption have an effect on the  perpetration of current partner abuse independent of other  significant main effects?".  Put another way, "is the influence of  stress and alcohol consumption dependent upon the effects of other independent measures included in the interaction terms?" Of interest  as well, was determining whether either of the interaction models  added to the explanatory power of the main effects model.  This  determination was made by contrasting the log likelihood ratios of  the main effects and interaction effects models.  The information  needed to assess the overall salience of these models is reported in the last row of Table 29.  The global tests of significance conducted  on the stress (L2=90.543, d.f. 23) and alcohol interaction models  (L2=73.160, d.f. 23) indicated that both were highly significant (p <  .0001).

The first contrast involved a comparison of the log likelihood ratios for the diathesis-stress main effects with the log likelihood ratios for the stress interaction model (Columns 1 and 2).  The second contrast compared the log likelihood ratios for the diathesis-stress main effects and alcohol consumption interaction model (Columns 1 and 3).

The nested comparison of the log likelihood ratio revealed that the terms included in stress interaction model improved upon the explanatory power of the main effects model.  The coefficients for the interaction terms as well as log likelihood ratio also suggest that there is a significant interaction between stress and underlying vulnerabilities as related to age and past perpetrated partner abuse in the prediction of current perpetrated partner abuse by males (X2=21.49, d.f.=4, p <.001).  According to the results, the effect of stress on the perpetration of current partner abuse is dependent upon whether male respondents were young and experienced episodes of past perpetrated partner abuse.

With respect to age,  males who experience high levels of stress and who are under the age of 49 years are more likely to perpetrate current partner abuse than males who are older or have low levels of stress.  The effects of past perpetrated partner abuse and stress on the perpetration of current partner abuse are such that males who experience high levels of stress and who have perpetrated partner abuse in the past are more vulnerable to current perpetrated partner abuse than those who do not have a history of past partner abuse under low or high stress conditions (See Figures 3 and 4).

Figure 3

     [Will be shown here when graphics file becomes available. --WHS]
 

Figure 4

     [Will be shown here when graphics file becomes available. --WHS]
 

The log likelihood ratio derived from a comparison of the  coefficients in diathesis-stress main effects model with the alcohol interaction model indicate little improvement by this interaction  model over the main effects model (X2=5.94, d.f.=4, n.s.).  No significant interactions emerged from this analysis.  This together with the inverse main effect for alcohol, suggests that the consumption of alcohol has little influence on the perpetration of current partner abuse on its own, or in combination with other variables.

The comparisons conducted demonstrate that the life stress events interaction model provides the best explanation of current perpetrated partner abuse by males.  Based on this model, life stress events' influence on current perpetrated partner abuse by males is demonstrated as a significant main effect as well as significant interaction effects.  When the main effect and interaction effects of  stress are compared, the former is associated with a high score on  the stress measure, whereas the latter is associated with the  opposite.

The antilogs show the net multiplicative effects of each  independent variable on the odds of perpetrating partner abuse.
     According to an examination of the antilogs emerging from the  logistic regression analyses conducted on the male data (based on  significant predictors from the main effects model), the following estimates emerged:

1. males who observed their fathers hitting their mothers have odds of perpetrating current partner abuse that are 4.569 times the odds of those who did not, other factors held constant,

2. being Catholic decreases the odds of males perpetrating current partner abuse by a factor of .030, other factors held constant,

3. for each increase in the Wave 1 CTS score (i.e., measuring past perpetrated partner abuse), the odds of males perpetrating current partner abuse is increased by a factor of 1.783, other factors held constant,

4. for each increase in the stress scale score, the odds of males perpetrating current partner abuse is increased by a factor of 1.696, other factors held constant,

5. for each additional year of age, the odds of males perpetrating current partner abuse is decreased by a factor of .906, other factors held constant.

Table 28. Coefficients representing the main effects of diathesis and stress measures and their interactions on the log odds of perpetrating current partner abuse among males based on Wave 1 and Wave 2 data

Note: Only predictors reaching a .05 level of significance or less are presented in this table.

Standard errors are reported under parameter estimates in parentheses.

* p < .05, ** p < .01, *** p < .001

(1) significant main effects for diathesis and stress variables

(2) stress interactions

(3) alcohol interactions

(4) odds ratios for significant main effects

Female data

Logistic regression analyses conducted on the female data followed the same approach as described for the male data.  The coefficients  in Column 1 indicate that of the 20 independent variables tested, the following four had statistically significant effects on the log odds of current partner abuse perpetrated by females: EPQP scores, Neuroticism Index scores, observing one's mother hit one's father and parents' mutual violence.  The parameter estimates show that females who had high scores on the EPQP and the Neuroticism Index, who  observed their mothers hitting their fathers and who did not observe their parents' mutual abuse were more likely to perpetrate current partner abuse than those who are not identified by these factors.
     The global test for the significance of the diathesis-stress main  effects model was L2=56.240 with 20 d.f. indicating that it is highly  significant (p < .0001).

As before, Columns 2 and 3 respectively report the parameter estimates of the stress and alcohol interaction models and assess whether each has an effect on current perpetrated partner abuse independent of other main effects.  As was done for the male data, the log likelihood ratios derived from the last row of Table 30 were used to assess the significance of each interaction model.  The contrasts conducted on the female data also followed the approach carried out on the male data.  The global tests for significance for the stress (L2=60.613, d.f. 23) and alcohol interaction models (L2=75.611, d.f. 23) were both highly significant (p < .0001).

The nested comparison of the log likelihood ratio revealed that the terms included in stress interaction model did not improve upon the explanatory power of the main effects model (X2=4.756, d.f. 3, n.s).
     Furthermore, no significant predictors emerged from the stress interaction model.  These analyses did not support stress' influence on current partner abuse perpetrated by females either as a main effect, or as an interaction effect, where its effect would be dependent upon the influence of other variables.

The log likelihood ratio derived from a comparison of the coefficients in diathesis-stress main effects model with the alcohol interaction model indicate that the terms included in the interaction model improved upon the explanatory power of the main effects model (X2=19.76, d.f. 3, p < .001).  The coefficients for the interaction terms suggest that there is a significant interaction between alcohol consumption and underlying vulnerabilities with respect to past perpetrated partner abuse, neuroticism index scores and observing mother hitting father in the prediction of current perpetrated partner abuse by females. According to the results, the effect of alcohol consumption on the perpetration of current partner abuse is dependent upon whether female respondents had a history of past partner abuse, had high scores on the neuroticism index, and observed their mothers hitting their fathers.

The interaction of alcohol consumption and these variables in predicting current perpetrated partner abuse by females have revealed some interesting associations.  Whereas the neuroticism by alcohol interaction demonstrated that the highest rates of current perpetrated partner abuse among females were found among those who consumed high levels of alcohol and who had high scores on the neuroticism index, the interactions involving past perpetrated partner abuse and observing mother hitting father presented somewhat different relationships.  Where females had neither perpetrated partner abuse in the past nor observed their mothers hitting their fathers, rates of current perpetrated partner abuse were very low regardless of the amount of alcohol consumed.  However, when there was a history of past perpetrated partner abuse or exposure to mother's violence, the highest rates of current perpetrated partner abuse were found among abstainers as well as high alcohol consumers (See Figures 5, 6, and 7).

Next: Chapter 5 Part 6