The purpose of this study was to determine what differentiates between anxious students who attend school regularly and anxious students with high absenteeism rates. We found that type of anxiety (degree of social anxiety and panic symptoms), the presence or absence of behavioural and substance abuse problems, psychiatric severity, perception of own health, and number of close friends all discriminated between these two groups. There were also differences with respect to two school factors, namely extent of bullying and whether or not students felt that they were treated with respect at school. These factors are interesting, because the anxious students who attended school regularly reported experiencing more bullying and feeling less respect at school than the anxious students with high absenteeism rates. This probably reflects the fact that the regular attendees spent more time in (what they regarded as) a hostile school environment than the high-absentee students did. The MDA analysis indicated that psychiatric severity and negative personality traits are the most important differentiating factors, followed by comorbid behavioural problems and the students' degree of resilience.
These findings may have important implications. First, they suggest the extent of anxiety-related problems in schools, and they indicate that negative personality traits and relational problems are the most important risk factors for school absenteeism. Function 1 indicates that it is the total symptom burden and the presence of negative personality traits that differentiate between the two high-anxiety groups. Other studies in recent years have reached similar conclusions. Psychiatric comorbidity, low socio-economic status, and employment have been identified as individual predictors for drop-out, and research also indicates that risk factors aggregate over time, eventually leading to premature departure from school [10–12, 34]. The findings from this and other studies indicate that some children may start school with individual risk factors that predispose them for absenteeism, such as anxiety, negative personality traits or negative parental attitudes towards school achievement, and school attendance. Over the years, developing psychopathology and/or learning disorders and an unsafe or poor school climate may further increase their chances for absenteeism. Function 1 also indicates that these students may have a tendency to interpret things in a negative way (they resent going to school, feel unsafe in school, and have a poorer perception of their own health). This may be a direct result of negative personality traits and may also reflect a tendency of high absentees to have more negative cognitions and to overgeneralize, as shown by Maric et al. .
There were several findings in this study regarding the difference between anxious attenders and non-attenders. The anxious school attenders were not as socially anxious as the anxious non-attenders, and they also had more friends. In addition, they reported being less frightened by the somatic symptoms of anxiety. These findings are supported by other studies [5, 8]. Heyne et al.  treated anxious school refusers with the @-School Programme. At follow-up, adolescents diagnosed with a social phobia had significantly lower school attendance than adolescents with another anxiety disorder and adolescents who no longer met the criteria for an anxiety disorder. They also found that lonely adolescents (those who reported having no friends in their class) were the worst at follow-up, mirroring findings from this study.
Furthermore, reactions to feelings of being scared or anxious seem to differentiate the two groups. The more students experience symptoms of panic, the more likely they are to be absent from school. Students who score high on this factor fear the sensations of anxiety and harm they may bring. Mattis and Ollendick  suggest that adolescents who react with panic symptoms have learned to associate negative events with physical symptoms and experience intense alarm reactions with little sense of predictability or control over stressors. This leads to apprehension and avoidance of situations that set off these alarm reactions—in this instance, any school-related issues or situations.
In this study, anxious school attenders had lower rates of behavioural problems than the anxious non-attenders. Both in the single risk factor analysis and in the MDA analysis, behavioural problems and related phenomena such as substance abuse were strong differentiating factors between the high- and low-absence high-anxiety groups. Function 2 in the MDA analysis indicates that students with more behavioural problems and substance abuse also experienced other phenomena typically associated with behavioural problems at a greater rate, such as negative life events and indicators of low socio-economic status (mother not working, mother with low education, and living alone). Egger et al.  described a group of school refusers who met the criteria for both anxiety disorders and disruptive behaviour disorders. Relative to the other groups in the study, this mixed group had higher rates of absence, was younger at age of onset, and was less active in extracurricular activities. They also had fewer friends, lower socio-economic status, and parents who were more likely to be treated for mental health problems. Thus, members of the mixed group came from home environments that lacked the conditions for a safe and secure upbringing.
There are some limitations in the present study that need to be considered. From these data, we cannot say anything about the temporal relationships between the risk factors and school attendance. This means that no causal inferences can be drawn from the data. However, the aim of this study was to examine the prevalence, characteristics, and differences between anxious school attenders and anxious school non-attenders. Second, using cut-off scores to define group membership can be viewed as a weakness. Cut-off points have been criticized for being arbitrary, with the chance that minor changes in the cut-off could lead to completely different prevalence rates and characteristics. We recognized this risk and took care to base our cut-off points for both anxiety  and absence  on empirically derived thresholds. Third, the study does not contain any measure of cognitive function in the students, and this may have affected the results. Although lower cognitive functioning has been associated with higher absenteeism in some studies, most studies have generally supported the notion that children with high absenteeism are of average intelligence and display adequate academic achievements prior to their absenteeism . Finally, the generalizability of the discriminant functions should be cross-validated to test the utility of the coefficients for other samples. This could be done by splitting a sample into two groups, deriving classifications in one group, and testing them on the other group. Another approach is to derive the classification functions from a sample at time one and retest them at time two . However, neither of these approaches was feasible in the present study.