CHAPTER 3: Methodology
3.7. Data analysis
The data was analysed using the Statistical Package for Social Scientists (SPSS, version 15.0).
The data was pre-coded for data-input into SPSS. Each questionnaire was numbered and the class and academic grading recorded from each class envelope. Each questionnaire was recorded separately. The data was first entered into Microsoft Excel as suggested by Tredoux
& Durrheim (2002) and then prepared and checked for accuracy. After this the data was transferred into SPSS and all analysis was completed in SPSS.
The purpose of the analysis was to establish the prevalence and forms of bullying with reference to the different bullying roles; to assess the degree of trauma experienced by
learners; and to determine the nature and direction of the relationship between the dimensions of bullying and symptoms of trauma experienced by learners.
3.7.1. Measures:
Once entered and audited, all data was assessed for validity as specified in the TSCC-A manual and by screening for missing data (Briere, 1996).
3.7.1.1. OBVS
This measure consisted of 38 items with the majority consisting of 5 point Likert-type scale items. Solberg & Olweus (2003) state that the optimal cut-off point to determine whether a learner is ‘involved’ or ‘non-involved’ in bullying is between “only once or twice” and “2 to 3 times a month” (p. 256). The distinction between bullies and victims is rendered by a
consideration of those learners who admitted to being a bully or victim and those who have experienced or have participated in bullying or victimisation behaviour. It was important that all frequencies of bullying be recorded as the analysis necessitated that the trauma T Scores of the TSCC-A subscales be compared against the frequency of bullying. As a result there are multiple definitions of bullying (based on frequency) that are described in the analysis.
Analysis focused on the various roles and frequency of bullying (bully, victim, bully/victim, and witness); the types of bullying; the locations of bullying; personal reactions to bullying and peer and teacher reactions.
3.7.1.2. TSCC-A
Each item was recorded on a 4 point Likert-scale ranging from 0 (never) to 3 (almost all the time) (Briere, 1996). Scale-scores for each of the 5 scales were calculated. The scores ranged from 0 to 27/30 (depending on the number of items in each scale) with higher scores
reflecting greater symptomology. The 5 subscale raw TSCC-A scores were then converted into T Scores based on the age and sex of the child. These were then checked for accuracy.
There were 486 questionnaires that could be accurately included according to the TSCC-A criteria. T Scores equal to or above 60 were considered sub-clinical and those above 65 were considered clinically significant (Briere, 1996).
The UND T and HYP T Scores were calculated and these scores were compared against the normative standards. A score of more than 70 for UND T is considered invalid for an individual clinical diagnosis, and those between 65 and 70 are viewed and interpreted with caution. HYP T scores greater than or equal to 90 are considered invalid, and those between 75 and 89 are viewed and interpreted with caution (Briere, 1996). There were 24 learners who were classified as UND and 11 learners who were classified as HYP. It was decided to
include these learners in the analysis, as the above distinctions are relevant to clinical intervention and do not have a negative impact in terms of the research questions under investigation. If anything, learners with UND responses are certainly not exaggerating their levels of trauma and learners with HYP responses might indeed be crying for help rather than
exaggerating their traumatic experience. In any event, the inclusion of these learners did not impact significantly on the direction and trends of the analysis.
3.7.2. Descriptive statistics
Descriptive statistics (frequencies, means, percentages and standards deviations) were used to analyse the biographical information (age, grade, academic grading and race) and the bullying and trauma measures. These statistics include the prevalence rates of bullying according to the specified roles (bully, victim-bully, victim, bystander). Frequencies, mean values and
standard deviations were used to facilitate a comparison between the various bullying roles and the TSCC-A subscales.
3.7.3. Inferential statistical analysis
Cronbach’s Alpha scores were calculated to determine the reliability of the TSCC-A
subscales. A correlation matrix was created between the forms of bullying and the five trauma scales. The Pearson product-moment correlation coefficient (r) was used to determine the degree of linear association between the variables, while coefficients of determination (R2) were calculated to determine the size of the effect. This correlational analysis thus indicated the direction and strength or degree of relationship between the variables. Correlations were also run between the biographical variables and the trauma scales and no significant findings emerged.
MANOVA’s which includes ANOVA’s were used to test for differences between the mean scores where there were 2 or more groups of independent variables. To determine whether significant differences exist between variables, such as the frequencies of bullying
experienced , a one-way multivariate analysis of variance (MANOVA) was conducted in order to assess the distribution of scores and determine how they differ from one another (Tredoux & Durrheim, 2002). Adequate cell and sample sizes are vital considerations in determining whether or not MANOVA is indicated. Post hoc tests (such as Levene’s test) were conducted to ensure that this type of analysis was relevant, with all underlying assumptions being carefully assessed to determine fitness of purpose.
Lastly, a binary-logistic regression analysis was run to test the predictive impact of bullying variables on sub-clinical and clinical traumatic diagnoses. The model was constituted by a number of predictor variables connected with the bullying experience, such as length of
bullying, whether the learner had skipped school, whether he was afraid of being bullied and the various types of bullying experienced. These were examined to determine the contribution of each variable in the model and the extent to which it accounted for the variance in trauma diagnosis. Although regression is a technique based on correlation, it enables a more
sophisticated multivariate exploration of the interrelationships amongst a set of variables than rendered by simple correlational analysis. Again, it needs to be highlighted that while this did not generate evidence of a causal relationship, it did indicate the variance in trauma diagnosis that can be explained by the predictor variables associated with bullying.