CHAPTER FIVE (part 1)

RESULTS OF WAVE 2 DATA

Examining the Data

Reliability of Scales

Prior to addressing the hypotheses set out in the previous chapter, two tests of reliability were performed on the independent and dependent measures in Wave 2 data.  The first assessed the degree of internal consistency of the measures using Cronbach's Alpha, and the second assessed the stability of the measures across time by way   of Test-Retest Reliability.  Lack of internal consistency as well as instability of measures limit the generalizability of research findings (Cronbach, Gleser, Nanda & Rajaratnam, 1972).

Internal consistency of the measures.  Cronbach's Alpha coefficients assess the proportion of variance due to common factors among scale items.  More specifically, the alpha coefficient is the ratio of the universe-score variance to actual observed score variance.  In other words, the alpha coefficient indicates how accurately one can generalize from an observed score with several modes of responses and a fixed situation, to the universe score for that situation over all modes of responses (Cronbach et al., 1972).

According to Kerlinger (1973), a reliability coefficient of .60 or better would be needed to meet a moderate standard of reliability.
     Listed below are the scale characteristics and reliability coefficients for measures put to this test.  The following measures were assessed using this technique: (1) the EPQ-R (EPQN, EPQP, EPQL, EPQE), (2) the Barron-Ego Strength Scale, (3) the MacAndrew Scale, (4) the Rosenberg Self-Esteem Scale, (5) the Spielberger Trait Anxiety Scale, (6) the Alcohol Dependence Data Schedule, and (7) the Conflict Tactics Scale (CTS). The remaining alcohol measures (Michigan Alcoholism Screening Test and the Diagnostic Interview  Schedule) were not subjected to tests of reliability since they provide multidimensional indicators of alcohol related behaviour.

1) The Eysenck Personality Questionnaire - Revised (EPQ-R)

The EPQ-R  is composed of the following four subscales: psychoticism (EPQP), neuroticism (EPQN), extraversion (EPQE) and a validity or lie scale  (EPQL).  For the purposes of these analyses, the long version of this  measure was employed.  Respondents were asked to answer "YES" or "NO" to all subscale items.

A. The EPQP is composed of 32 items.  The scale range was 0-14 for males and 0-12.38 for females with means of 3.78 (S.D.= 2.59) and 3.25 (2.39), respectively.  The EPQP provided Alpha levels of .58 for males and .54 for females.

B. The EPQN is composed of 24 items.  The scale range was 0-24 for both males and females with means of 7.76 (S.D.=5.16) and 10.17 (S.D.=5.25), respectively.  The EPQN provided Alpha levels of .87 for both males and females.

C. The EPQE is composed of 22 items.  The scale range was 0-23 for both males and females with means of 13.53 (S.D.=5.29) and 13.45 (S.D.=4.63), respectively.  The EPQE provided Alpha levels of .86 for males and .82 for females.

D. The EPQL is also considered to be a measure of social conformity (Eysenck & Eysenck, 1985).  It is composed of 21 items. The scale range was 0-20 for males and 0-21 for females with means of 9.33 (S.D.=4.48) and 10.48 (S.D.=4.49), respectively.  The EPQL provided Alpha levels of.82 for males and .83 for females.

2. Barron Ego Strength Scale

The Barron Ego Strength Scale is composed of 67 items.  Respondents were asked to answer "TRUE" or "FALSE" to each of the scale items. The scale range was 25-58 for males and 24-59 for females with means  of 47.41 (S.D.=5.41) and 43.85 (S.D.=5.68), respectively.  The Barron Ego Strength Scale provided Alpha levels of .63 for males and .60 for females.

3. The MacAndrew Scale

The MacAndrew Scale (MAC) is composed of 49 items. As with the Barron Ego Strength Scale, respondents were asked to answer "TRUE" "FALSE" to the scale items. The scale range was 10-34 for males and 10.42-32 for females with means of 22.05 (S.D.= 3.80) and 20.05 (S.D.=3.49), respectively. The MacAndrew Scale provided Alpha levels of .33 for males and .25 for females.

4. Rosenberg Self-Esteem Scale

The Rosenberg Self-Esteem Scale is composed of 10 items.  Respondents were asked to indicate the extent to which they agreed or disagreed with each scale item.  Choices of responses were: (1) strongly agree, (2) agree, (3) disagree, and (4) strongly disagree.  When computed, high scores indicate high self esteem, whereas low scores indicate the opposite.  The scale range was 20-40 for both males and females with means of 33.83 (S.D.=4.50) and 33.30 (S.D.=4.49), respectively.
     The Rosenberg Self-Esteem Scale provided Alpha levels of .85 for both males and females.

5. Spielberger Trait Anxiety Scale

The Spielberger Trait Anxiety Scale is composed of 20 items. Respondents were asked to indicate the extent to which they  experienced scale items.  As was the case with the Rosenberg Self-Esteem Scale, this measure was also based on a four point scale.
     The item values were as follows: (1) almost never, (2) sometimes, (3) often, and (4) almost always.  The scale range was 20-61 for males and 21-71 for females with means of 32.88 (S.D.=7.86) and 34.72  (S.D.=8.20), respectively.  The Spielberger Trait Anxiety Scale  provided Alpha levels of .88 for males and .89 for females.

6. Raistrick's Alcohol Dependence Data Schedule (SADD)

The SADD is composed of 15 items.  As in the previous scale, respondents were asked to indicate the extent to which they experienced the scale items. Scale values were similar to those of the Trait Anxiety Scale and were as follows: (1) never, (2) sometimes, (3) often, and (4) nearly always.  The scale range was 0-13 for males and 0-12 for females with means of 1.15 (S.D.=2.30) and .96 (S.D.=2.02), respectively.  The SADD provided Alpha levels of .76 for males and .66 for females.

7. Conflict Tactics Scale (CTS)

The abridged version of CTS used in this research is composed of 6 items reflecting the more severe forms of physical abuse. Respondents were asked to indicate how often they participated in the various forms of conflict resolution strategies reflected by the scale items.  The CTS was constructed on the following six point scale: (1) never, (2) once a year, (3) two to three times a year, (4) often, but less than once a month, (5) about once a month, and (6) more than once a month.  The scale range was 6-22 for both males and females with means of 6.44 (S.D.=1.53) and 6.70 (S.D.=1.66), respectively.  The CTS provided Alpha levels of .83 for males and .74 females.

Table 6 provides a summary of Wave 1 and Wave 2 reliability coefficients for male and female respondents.  Upon comparing the Alpha levels from both sets of data, it appears that for the most part, estimates of internal consistency remained relatively stable across the two year period between data collection for both males and females.  The exceptions are the reliability coefficients provided by the MacAndrew Scale (male and female data), the Ego-strength Scale (female data), the SADD Scale (male and female data) and the CTS (female data).  The source of these differences will be explored in an examination of attrition that follow in a later section.

Table 6. Cronbach's alpha coefficients for male and female respondents in Wave 1 and Wave 2.

Test-retest reliability of the measures

One of the objectives of this study is to examine the stability of the partner abuse across   time.  In order to better understand this variable, the stability of   other independent measures also needs to be explored.  Wave 1 and Wave 2 measures were correlated and assessed by Pearson's correlation coefficients to determine their test-retest reliability.  Of the 13 measures tested, six for males, and five for females attained r  values of .70 or greater. In general, correlations were stronger for males compared to females.

As expected, personality measures were found to be more stable than alcohol or spouse abuse measures.  Strongest correlations were provided by the EPQE (r=.86) for males, and the EPQL (r=.80) and EPQN (r=.80) for females.  Weakest correlations were provided by the MAST and the "lifetime diagnosis for alcoholism for both males (r=.47 and r=.17, respectively) and females (r=.27 and r=.11, respectively).
     Correlations for the prevalence of partner abuse (CTS) were .59 for  males and .44 for females. Table 7 summarizes the results of these analyses based on male and female respondents who completed  questionnaires in both Wave 1 and Wave 2.

Table 7. Test-Retest Reliabilities: Pearson Correlation Coefficients on Wave 1 and Wave 2 measures for male and female respondents

Note: All correlations are significant at the p < .001 level except for female's lifetime diagnosis for alcoholism (p < .05).

Rates of Attrition

It has already been reported that an attrition rate of 21.3 percent has been experienced by the entire sample.  Of the subsample of males and females who were married or remarried, this rate was found to be 20.7 percent (21% for males and 20.4% for females).

T-Tests and Chi-Square analyses were conducted within the married and remarried subsample to assess whether systematic differences existed between respondents who completed Wave 2 of this project and those who did not.  Both male and female dropouts had significantly higher MAST scores than male and female completers (1.36 v. 1.00 , p < .05 for males, and .58 v. .37,p < .001 for females).
     Male dropouts alone differed from male completers along the following dimensions: male dropouts tended to be nonwhite (46.33%, p < .01), belonged to the religious preference category, "other" (35.39%, p < .05), had higher EPQP scores (4.66 v. 3.63, p < .01), and consumed more alcohol (.74 ounces v. .55 ounces, p < .001).
     Female dropouts on the other hand, were significantly different from female completers in that they tended to be older (28.66%, p < .01), had higher SADD scores (1.29 v. .79, p < .001), consumed less alcohol  (.19 ounces v. .28 ounces, p <.001) and had higher scores on the "lifetime diagnosis for alcoholism measure" (1.30 v. 1.08, p < .05).

T-Tests conducted on CTS mean scores (as measured by the abridged version of the CTS) did not produce any significant differences between dropouts and completers for either male or female respondents.  Similarly, the proportion of male and female respondents reporting perpetrating partner abuse did not differ significantly for either completers or dropouts.  Tables 8 and 9 provide the results of chi-square and t-test analyses conducted on Wave 1 demographic, personality, alcohol, and the prevalence of perpetrated partner abuse with respect to participation in this project.

Table 8:  Sample attrition by demographic and partner abuse variables for married and remarried, but previously divorced male and female respondents who participated in Wave 1

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

Table 9.  Personality, alcohol and partner abuse scores by study participation for married and remarried male and female respondents.

Note: * p < .05, ** p < .01, *** p < .001
      Not all totals will equal 94 (males) or 92 (females) for dropouts or 352 (males) or 360 (females) completers due to missing data.

Next: Chapter 5 Part 2