Introduction

Bullying among children and adolescents has been a public concern for many years. Cross-sectional [1] and longitudinal studies [2] show that being a victim of bullying is associated with, e.g., anxiety, depression, suicidal ideation, lack of social relationships, economic hardship, and effects on education in adulthood. Previous studies using the 36-item Short-Form Health Survey (SF-36), covering eight health domains, reported that bullying was associated with worse health-related quality of life (HRQoL) across all health dimensions [3, 4]. The prevalence rates of bullying vary from a few percentages to over 30 % in some countries [5]. The lowest prevalence rates are often reported in Scandinavian countries, and in Sweden, 1–7 % of 11- to 15-year-old pupils report being exposed to bullying [6].

A large number of different bullying prevention programs have been implemented in schools around the world in the last decades in an attempt to reduce the prevalence of bullying. However, few bullying prevention programs have been evaluated by studies using a credible design for causal inference—that is, experimental or valid quasi-experimental approaches [7, 8]. One meta-analysis focusing on specific components of prevention programs, rather than specific named programs per se, showed that well-designed prevention programs may reduce the prevalence rate by approximately 20 % [9]. Prevention programs obviously come at a cost, and economic evaluations based on cost-effectiveness analyses (CEA) are increasingly being used by decision makers in health policy jurisdictions as inputs in the decision-making process. The Swedish National Agency for Education [10] has estimated the economic cost for a number of bullying prevention programs, showing that the cost per pupil and year ranges from 279 Swedish kronor for the School Comet program (€30) to 1389 Swedish kronor for the Olweus Bullying Prevention Program (€150). But, as far as the authors know, there have not yet been any studies that analyze whether these programs constitute a cost-effective use of resources—that is, that relate costs to a relevant outcome measure. The recommendation for economic evaluations of health and social services by, for example, the US Public Health Service Panel on Cost-Effectiveness, the UK National Institute for Health and Clinical Excellence (NICE), and the Swedish National Board of Health and Welfare, is that cost-effectiveness studies should compare the cost of a new intervention to the gain in quality-adjusted life-years (QALYs), a form of CEA often called cost-utility analysis (CUA) [11–13]. QALY is an outcome measure that combines the length of life (life-years) with preference-based health-related quality of life into a single measure that makes it comparable across different health intervention areas [11, 14, 15]. The quality of each life-year is assessed with a preference-based health-related quality of life estimate (sometimes called “utility score” or “QALY weight,” henceforth in this paper referred to as utility score) between 0 and 1, where 0 is scaled to be “equal to death” and 1 is interpreted as “perfect health.”

Today, there are a large number of studies looking at health state utility scores for general populations as well as for specific health conditions. For example, van den Berg [16] showed that the utility score in the UK general population decreases as age increases and that 16- to 19-year-old adolescents had a mean score of 0.861 for males and 0.829 for females. Similar utility scores were found for Swedish youth (0.89 among ages 20–29) [17]. However, there are markedly fewer studies on disorders and specific life consequences primarily affecting children and adolescents and as far as we know no studies involving individuals who are victims of bullying, which is related to a number of different health consequences (see, e.g., discussions in [18]; [19]; [20]).

In order to conduct economic evaluations in the form of cost-utility analyses of bullying prevention programs, it is thus necessary to estimate the utility score for victims of bullying compared to non-victims. Therefore, the aim of this study is to assess the utility score among victims of bullying compared to non-victims. The results can be used to compare the burden of bullying compared to other health and social problems as well as being used as an indicative input in cost-utility analyses of bullying prevention programs. For the latter, it is essential to estimate the causal effect of bullying on utility scores, which is not something we have the possibility to do with our data. The hypothesis is that victims of bullying have a lower rate of health-related quality of life compared to non-victims of bullying.

Methods

Survey design

This study is based on data from the third wave of a Swedish longitudinal research project investigating different aspects of the psychosocial functioning of children and adolescents, including their experience and perception of bullying. The study was carried out in Gothenburg, Sweden’s second largest city, with a population of approximately 550,000 (in 2010). These data have been used to report adolescents’ HRQoL and its association with bullying [3], but not reporting preference-based HRQoL estimates.

Participants and procedure

In total, 874 pupils were invited to participate by post. A total of 758 (429 girls) adolescents aged 15–18 years (M = 16.2, SD = 0.5, almost all aged 16 or 17) from different schools and types of neighborhood in Gothenburg participated and constitute the sample in this study. From the age of 6, all children in Sweden can be admitted to preschool class. Between ages 6/7 and 15/16, children attend compulsory comprehensive school, and from 16/17 children attend high school, which is voluntary, but only 2 % chose not to. The vast majority of schools in Sweden are municipally run, but there are also autonomous and publicly funded schools, known as “independent schools” (SNAE 2015).

Enclosed with the invitation were a questionnaire and a prepaid response envelope. The pupils were offered a movie ticket as compensation once their questionnaire had been sent back. After 2 weeks, a reminder including another copy of the letter was sent to those who had not returned the questionnaire.

Measures of health and bullying

SF-36 and SF-6D

Adolescents in the survey were given the SF-36 health questionnaire (version 1), which is a generic 36-item multi-purpose tool for evaluating individuals’ health [21]. It includes a multi-item scale that assesses eight different major health domains: (1) physical functioning, (2) role limitations due to physical problems, (3) social functioning, (4) bodily pain, (5) general mental health, (6) role limitations due to emotional problems, (7) vitality, and (8) general health perceptions. Each domain is scored on a scale ranging from 0 to an optimal result of 100, but the scores do not facilitate comparisons across dimensions.

To address the concerns around comparability and how individuals’ judge the importance of different domains, a scoring algorithm is used to translate the SF-36 to SF-6D [22]. SF-6D is based on all dimensions in the SF-36 apart from the question on respondents’ general health perception. In the original analyses, 249 different health states in the SF-6D were valued by a random sample of the population, allowing the prediction of a utility score for each of the 18,000 possible health states [22]. The resultant SF-6D index is scaled from 0.301, which is the worst health state, to 1, which is the best health state. Today, the SF-6D is used extensively to estimate utility scores for cost-effectiveness studies, as are (for example) the EQ-5D and Health Utility Index (HUI) [22–24].

Bullying

The questions involving bullying were based on the standard definition of bullying, which emphasizes its repetitive nature and the power imbalance between the perpetrator and the victim [25, 26]. Self-reported bullying was assessed by the victimization index [27], which asks adolescents to indicate how often they have been bullied at school during the current year. The possible responses were as follows: “every day,” “most days,” “1 or 2 days a week,” “about once a week,” “less than once a week,” and “never.” Students were classified as victims if they had answered that they had been bullied at least one or two days a week or more frequently during the current year. The pupils were also asked about bullying in previous years—that is, how often they had been bullied during their time in lower secondary school (7th, 8th, and 9th grade, i.e., 13, 14, and 15 years old, respectively) [for further details on this, see 3].

Background/control variables

Data on the age of the respondent, gender (boy/girl), and country of birth for students (“In which country were you born?”) and for parents (“In which country was your mother/father born?”) were also collected based on self-reports. Country of birth was dichotomized into “born in Sweden” or “born in other countries.”

Analysis

We compared the utility score with the adolescents’ bullying experience in the first year of high school (which they currently attend) and, retrospectively, grade 9 (previous school year). It should be mentioned that there is in general a strong correlation in the bullying experience over grades—that is, the adolescents being bullied in high school are also very likely to have been bullied in grade 9, etc.

Initially, we used summary statistics to describe the prevalence of bullying, as well as the mean utility score for adolescents who were never victims of bullying, bullied in high school, bullied in grade 9, or bullied in both high school and grade 9. We tested for statistical significant differences using parametric (t test) and nonparametric (Mann–Whitney U tests).

We also conducted multivariate linear regression analyses with the utility score as the dependent variable and the different bullying experience as independent variables. The analyses were run both excluding and including the control variables, such as age and gender, if the adolescent was born in Sweden, and if their parents were born in Sweden. All analyses were conducted using the software Stata version 13.

Moreover, since the dependent variable is truncated between 0.3 and 1 and there is usually a spike of observations close to 1 (i.e., many respondents have very high HRQoL), there are reasons to suspect that ordinary least squares (OLS) may give biased results [28]. Thus, we also conducted all our analyses with a regression estimator based on the beta distribution, which takes the specific features of HRQoL data mentioned above into account. These estimations resulted in identical results in terms of the effect of being bullied on HRQoL, and we therefore kept to the simpler OLS models in the presentation of our results in the paper.

Results

Table 1 shows bullying prevalence rates in the adolescent population. The rates for being a victim of bullying in the current year (only) (first year in high school/upper secondary school) as well as the reported response on whether or not they were bullied in the previous school grade (grade 9; final grade in lower secondary school) are presented. Finally, we also show the prevalence rates for those adolescents who report being bullied during both the current and previous school year (high school and grade 9).

Table 1 Descriptive statistics on prevalence of bullying victimization (%)
Full size table

The prevalence rates indicate that bullying is markedly less of a problem in high school than in grade 9 (3.19 % compared to 9.73 %). In our sample, more girls than boys are exposed to bullying in grade 9, but more boys are exposed in high school. However, performing statistical tests shows that the differences in reported prevalence for boys and girls are not statistically significant at the 5 % level. Thus, we cannot reject the null hypothesis that there is no difference in the prevalence of being a bullying victim between boys and girls.

The main variable of interest in the paper is the utility score. In the full sample, it varies from 0.419 to 0.965 with a sample mean of 0.758 and a median of 0.773, and in Table 2, we show descriptive statistics on the utility score in the different bullying experience categories.

Table 2 Descriptive statistics on mean preference-based health-related quality of life score (95 % CI)
Full size table

The results show that the mean utility score is 0.76 for adolescents who have not been bullied in grade 9 or high school. Adolescents who report being bullied report a mean score of 0.71 (this or the previous year), which is statistically significantly different compared to 0.76 (p < 0.001). The score is even lower, at 0.69, for those adolescents bullied in both the current and previous year. The latter estimate is based on quite a small number of adolescents, and therefore, the 95 % CI is rather large (but the negative effect is still statistically significant with p < 0.001). Due to the relatively small sample and the few number of pupils being bullied, we do not separate the analyses based on gender.

In Table 3, we report the results from three different regression analyses. What separates the three models from each other is the definition of the exposure variable: We report results using bullying variables that include adolescents being bullied this year (model 1), bullied last year (model 2), and bullied in both the current and previous year (model 3). We also control for age, gender, immigrant status, and parents’ immigrant status in all three models (each model is analyzed with and without the control variables).

Table 3 Regression estimates on health-related quality of life (HRQOL) score
Full size table

The results show that being bullied, compared to not being bullied, was associated with significantly lower utility scores in all models and with bullying in all grades. Being bullied in high school compared to not being bullied is associated with a lower score of 0.05–0.06 points (p < 0.05). The same result is found for adolescents being bullied in grade 9. With the group of adolescents being bullied in both grade 9 and high school, the utility score is 0.07–0.08 points lower than 0.76. As for the control variables, we see that they add some explanatory power to the variation in the utility score, and specifically so controlling for gender and fathers’ immigrants status. Girls and adolescents with an immigrant father have a lower mean utility score.

Discussion

This study is, to the authors’ knowledge, the first to estimate utility scores (preference-based HRQoL) for adolescents being bullied compared to non-bullied adolescents. In line with our hypothesis, mean utility scores were higher (0.76) for non-victims compared to victims (0.69–0.71) of bullying. The difference in utility scores is thus larger than what is often referred to as the minimally important difference for evaluative purposes of 0.03 [20, 29]. The result can also be interpreted as each year the adolescents who are victims of bullying lose 0.06 life-years compared to non-victims. It should be emphasized that this is the difference in utility scores for adolescents currently (or last year) being bullied. Previous research has indicated that bullying is also linked to several long-term negative health effects, e.g., increased risk of depression in later adolescence and adult mental illness [2, 30, 31]. In studies evaluating anti-bullying interventions, it may thus be necessary to consider both the short-term and long-term consequences of bullying.

The reason why some children are being bullied over several years is not clear. But we know that individuals with behavior deviant from the peer group are more at risk of being bullied. This may explain why adolescents, even when changing school as from grade 9 to high school, still are being bullied. There may also be an individual factor affecting the experience of being bullied. Some children may be more sensitive than others to their surroundings and peers’ expressions and also more sensitive to mental health problems.

Utility scores are often used in cost-effectiveness evaluations in terms of comparing the (incremental) cost to the gain in quality-adjusted life-years (QALY = utility score × life length) for an intervention compared to an alternative intervention/status quo. Most frequently, this is used in evaluations of healthcare treatments [32]. However, it is increasingly being used for evaluation broader public health interventions [33] and is also used in, e.g., social care settings, such as evaluating health effects of social housing [34]. But we may consider other types of economic evaluation approaches in the bullying context, such as cost–benefit analysis, to try to encompass all health and non-health consequences [35].

In this study, we used the SF-6D to estimate utility scores, and we may have ended up with slightly different results if we would have been able to use, for example, EQ-5D or HUI to estimate utility scores. The SF-6D is sometimes argued to have considerable flooring effects, whereas it is often argued to be relatively better, in discriminating between good health states. For example, a recent study evaluating a large number of multi-attribute utility instruments found that 1.3 % of individuals scored 1.00 based on the SF-6D, whereas 19 % of the same sample of individuals scored 1.00 with the EQ-5D [36], which is the most commonly used instrument for estimating utility scores. In this study, we are analyzing mostly (relatively) healthy and young individuals, which makes it more appropriate to have better discriminatory effect for good health states.

We found that the utility score of non-victims differed (utility score 0.76) rather substantially from normative population preference-based HRQoL reported in the group of 16- to 19-year-olds (mean score of 0.861 for males and 0.829 for females; van den Berg, 2012). Our results were more similar to the older populations’ scores in the van den Berg study (2012), ranging from 0.79 among 45–49 years old to 0.70 among 80–84 years old. One possible explanation to these differences may be cohort trends changing over time or cultural differences. Another explanation may be that different methods to collect data have been used in the studies and thus affected the results. In order to control for such changes, there is a need for additional samples to compare with. Our results also show girls having a lower utility score than boys, which is in line with previous studies [17, 37].

Having a mental health problem, internalizing or externalizing but in particular hyperactivity disorder, increases the risk of being bullied. This could perhaps be due to behavior that irritates or provokes other children [38]. Being bullied could also cause mental health problems such as depression, anxiety, and suicidal ideations [2]. Therefore, depression, depressive symptoms, and hyperactivity disorders may be appropriate comparison conditions, given that they overlap to meaningful degree with bullying. In the norm score population presented by van den Berg (2012), clinical depression rated a mean preference-based HRQoL score of 0.641 (0.633–0.648) among adults. Further, Sullivan and Ghushchyan [39] reported a slightly higher—and closer to our scores—mean score of 0.732 for depressive symptoms (in a population with a mean age of 48). However, it must be recognized that these are scores for an adult population and scores for adolescents may not be the same.

Limitations

The main limitation of the study is that we have estimated the association between utility score and being a victim of bullying—that is, we cannot tell whether the lower utility is the consequence of being bullied or caused by the fact that adolescents with several health (and/or other) problems are more likely to be bullied. Our only possibility for addressing these concerns was to include a number of individual characteristics in the regression analyses. We could also have used techniques such as propensity score matching, but also such an approach relies on selection on observables and hence does not properly address our concern, i.e., that unobservable factors are related to both lower utility and bullying, as well as potential reverse causality. However, it seems likely that a potential bias in our results would imply that we have overestimated the negative consequences of bullying (the different potential mechanisms for bias mentioned above would reduce the estimated difference in utility score). Hence, in cost-utility analyses of bullying prevention programs we can interpret the difference of 0.06 as an upper bound of the “true” causal effect of being a victim of bullying on preference-based HRQoL. In order to be considered a worthwhile investment, an economic evaluation of a bullying prevention program should be significantly cost-efficient using the utility score difference of 0.06 between victims and non-victims (given the possible overestimation of the difference), and sensitivity analyses should be carried out using smaller utility score differences as well.

Additionally, one limitation is that we used preference-based weights for the SF-6D based on the UK population. Unfortunately, there are no preference weights for the Swedish population for the SF-6D. However, the UK weights have been used extensively in studies using Swedish data. Further, given that we are mainly concerned about differences in the utility score between victims and non-victims of bullying (and not absolute levels), the potential bias from using UK weights may be less problematic. Finally, it is important to underline that the conclusions can only be generalized to adolescents in urban areas in Sweden, because the study was conducted only in one city of the country. It is possible that the results can differ from rural area.

Conclusions

We estimated preference-based HRQoL (utility score) for adolescents being bullied compared to non-bullied adolescents, and mean utility scores were 0.76 for non-victims and 0.70 (0.69–0.71) for victims of bullying, respectively. Considering both potential reverse causality and unobservable factors that may cause both low utility and bullying, it is important to note that the estimated difference in utility score cannot be interpreted as the causal effect of being bullied on utility. But, given the probable sign of the bias (overstating the true causal effect), we interpreted our results such that in cost-utility analyses of bullying prevention programs the difference of 0.06 can be used as an upper bound of the “true” causal effect of being a victim of bullying on preference-based HRQoL. In future research, it is desirable with bigger datasets that can control for more potential confounders and/or use quasi-experimental approaches to reduce the potential risks for biased estimates as discussed in our study.