Sticks and Stones…and Mobile Phones? Bullying is Prevalent in Our Schools

The Problem of Cyber Bullying

Teachers and parents are still learning about the new forms bullying can take and why they can be just as harmful as traditional forms. Some of the main problems with cyber bullying include:  

• Wide Audience
• Permanence
• Anonymity 

Aims of the Research

• To assess the prevalence of various forms of bullying, including cyber bullying; 
• To identify the frequency and circumstances around different bullying behaviours;
• To gain an understanding of attitudes and behaviours of teachers and students, and characteristics of schools in relation to bullying; and
• To investigate psychological and social factors associated with bullying. 

Challenges

The survey presented certain challenges requiring a sophisticated mathematical approach.

• A complex sampling approach to ensure the best national representation of students and schools with the limited sample size available.
• Weighting across the sampling factors needed to be done to make meaningful inferences about the entire Australian population and sub-populations of interest. 
• Bullying tends to be nested within classes and schools leading to a lack of statistical independence which needs to be taken into account when making statements about the precision of the findings. 

The Data Analysis Australia Approach

Sampling 

The sampling design created by Data Analysis Australia was necessarily complex as the population which it had to represent was complex – students within classes within years within schools within sectors within states. Based on the initial suggestions from the ECU team, a stratified, multi stage method was developed to balance the needs of being representative and allowing inferences to be drawn for many sub-populations. 

The final strata were based on state, area, sector (government/non-government) and level (primary/secondary) with a small number of schools being randomly selected within each stratum. Once a school was selected, all students in the relevant year groups were invited to participate in the survey. 

Importance of Weighting

To enable meaningful interpretation to be applied to any survey data, survey weights need to be calculated to reflect the sampling.

•  An important feature of the weighting process is that it can reduce bias by compensating for unequal sampling probabilities, which were introduced to ensure adequate representation of all strata.
• Data Analysis Australia took into account both the probability of a school being selected and the probability of a student responding, given that their school was selected.
• Weights should be considered as expansion factors permitting the scaling of the sample to the population, allowing inferences to be made about the entire population from which the sample is drawn. 

Precision of Results

Any survey that does not sample everyone in the population is going to be subject to error. This by no means invalidates the results of the survey – it is just important to understand the level of precision or “confidence” that can be placed on the results. Confidence intervals are a common method of achieving this, by giving not just a single estimate, but a range in which we can expect the true result to be within, with some level of confidence (typically 95%). 

The process that Data Analysis Australia employed had to handle not only the complex sampling process and weighting of the data, but also the non-independence of the data which results from the nesting of bullying within schools. Failure to properly take this non-independence into account would give inflated levels of confidence in the results. 

The computation of confidence intervals was done using bootstrapping.

• The bootstrap is a re-sampling technique whereby samples are simulated from the actual data to match the variability that would be observed if multiple real samples were used.
• Thousands of resampling simulations can be conducted, following the same constraints of the original sampling. The estimate can then be recalculated for each of these simulated samples to provide a simulated (or empirical) distribution of the estimate.
• The confidence interval is then defined by the appropriate quantiles of this simulated distribution. For example, the 95% confidence intervals for the estimates were determined by taking the 2.5^th and 97.5^th percentiles of the distributions. 

Testing for Differences in Attitudinal Behaviour 

An important aspect of the research was to analyse the attitudes and behaviours of students in relation to bullying and investigate psychological and social factors associated with bullying. Factors to be assessed included depression, loneliness, social, emotional and behavioural difficulties, and a sense of connectedness to their school. 

Both linear regression and logistic regression were performed to test for significant differences between students who were cyber bullied and students who were not. This method enabled the effects of age, gender and other factors to be taken into account to separate out the remaining differences between those being bullied from those not being bullied and between those who bullied from those who did not bully. These differences included the student’s:

• Level of connectedness to their school;
• Loneliness at school;
• Depression (based on an established psychometric scale);
• Whether they stayed away from school; and
• Whether they felt safe at school. 

The Result

The analysis has provided prevalence estimates and extensive information about covert and cyber bullying in Australian schools. The release of the Australian Covert Bullying Prevalence Study report was announced by the Federal Government on 1st June 2009, and its findings have highlighted the extent of the problem and the impact covert and cyber bullying has on children and adolescents.

July 2009