A mood-related ruminative response style refers to how a person, when dysphoric, focuses attention on his or her symptoms, and their ‘potential causes, implications and consequences’ . Rumination is frequently studied alongside affective psychopathology and is usually assessed using the multi-item Response Styles Questionnaire (RSQ) . High rumination scores on the RSQ predicts higher future depressive symptoms  and DSM-defined major depressive episodes in child and adolescent samples [4–6].
This rumination-depression association appears to be stronger in studies of adolescents than of children . This may be due to greater exposure to negative stressors from the age of 13; this is relevant because rumination may moderate the depressogenic effect of stressors . Alternatively this may reflect differences in other cognitive vulnerability factors that manifest differentially with increasing age . Another possible explanation is that it is the effects of puberty (with the change in hormonal milieu) that increases the depressogenic effect of rumination, rather than age itself. This has not been tested to date.
Potential confounders to rumination-depression associations
Self-rated rumination is strongly correlated with concurrent depressive  and anxiety [8–10] symptoms, which are themselves strong predictors of future depressive symptoms and disorder . This may confound the associations found in above studies between rumination and future depression. It has been argued that this is partly because high levels of depressive symptoms would themselves make scores on some rumination items higher. For example a score on the RSQ item ‘I think about how sad I feel’ may be high either because of a high tendency to ruminate on low mood, or the fact that the person is currently very sad so thinks about this a lot . Therefore high RSQ scores may be associated with future onset of depression because high concurrent depressive symptoms lead to both high RSQ scores and high risk of depression; and it may be the case that the cognitive style of ruminating has no effect on depressive symptoms.
Two methods have been used to control for such potential depressive symptom –rumination confounding. Firstly, some studies have statistically controlled for baseline depressive symptoms. For example, controlling for depressive symptoms attenuated the correlation between baseline rumination and follow-up depressive symptoms from r = 0.3 to r = 0.07 (95% CI 0.03-0.11) in a meta-analysis of childhood/adolescence studies . Two studies in adolescents have found that rumination scores are associated with future onset of depressive disorder, even when controlling for concurrent depressive symptoms [4, 6]. No studies have controlled for prior levels of anxiety in addition.
The second method has been to restrict use of items to those from the rumination questionnaire that are likely to measure actual rumination, as opposed to items that are strongly influenced by current depressive symptoms. Often such studies have used linear factor analysis methods or principal components analysis (PCA) to explore multiple dimensions among item sets. Initial studies in community-recruited adults [9, 11, 12] identified a ‘brooding’ factor/principal component, which was more strongly associated with current and/or future depressive symptoms than any other dimension of the RSQ (in particular a ‘reflecting’ factor/PC). However, while some studies in adolescents found a similar two factor structure of the RSQ [10, 13], one found only a single factor .
In this study a third methodological approach is considered that might better separate the rumination construct from depressive and anxiety symptoms. If some items from the rumination questionnaire are in fact measures of depressive symptoms, they would be expected to correlate strongly with items from questionnaires measuring depressive symptoms. If items from both the rumination and depressive questionnaires were entered into the same factor analysis, we could identify whether such items (ie ‘depression’ items from the rumination questionnaire) load better with the depressive symptom items, the rumination items, or are in fact part of a separate construct. Likewise, the addition of items from an anxiety questionnaire would identify rumination questionnaire items that load better with anxiety symptoms. It is also possible that items from all questionnaires would inter-correlate strongly with each other, and the factor analysis would suggest just a single common factor as a parsimonious solution. In this case, items would be best seen as measuring one common construct; this construct could be termed ‘negative cognitions’, and would be a risk factor for future depression (and possibly anxiety). Studies to date have made the prior assumption that items from rumination, depressive symptom and anxiety symptom questionnaires measure separate constructs, so should more appropiately be analysed separately. We propose that entering all items into a pooled factor analysis could explore whether this assumption is likely to be correct.
Distraction and problem-solving
In addition to the features already discussed, the RSQ also contains two further sub-scales called distraction and problem-solving, which are thought to be adaptive responses to low mood. Factor analysis suggests that items from both scales load onto a single factor [15, 16].
High levels of distraction and problem-solving have been found to be associated with reduced future depressive symptoms in community  and high-risk  samples of children and adolescents, controlling for prior depressive symptoms. In addition, the ratio of rumination to distraction/problem-solving is associated with increased depressive symptoms at follow-up, suggesting that high levels of distraction/problem-solving mitigate some of the effects of rumination [15, 16]. As high distraction/problem-solving in itself probably leads to reduced depression risk (rather than just reducing the effects of high rumination), a linear ratio approach has been considered a better way to model this data than a rumination x distraction interaction .
Goals of the current study
Our analysis consisted of two phases. In the first phase, we investigated the factor structure of a joint set of items from all three self report questionnaires purporting to measure depressive symptoms, anxiety symptoms and rumination. We hypothesized that this would either identify all items measure one underlying ‘negative cognitions’ construct; or alternatively identify multiple separate constructs, including rumination (and possibly different forms of rumination). This factor analysis would ideally assign each item from the pooled item set to the appropriate construct.
In our second phase, we hypothesized that high rumination would predict onset of a depressive episode over the subsequent 12 months and high depressive symptoms 12 months after baseline; this would be true controlling for confounding from baseline depressive and anxiety symptoms. As the first phase would identify whether RDQ (rumination questionnaire) scale items loaded best with the rest of the RDQ questionnaire or other questionnaires, the a priori plan was to only include scores from items found to load with ‘rumination’ to make up this total rumination scale for this analysis. We also hypothesized that a high ratio of rumination to distracting/problem-solving response styles would be associated with a higher risk of depression onset/symptoms, indicating that these adaptive response styles partially mitigate the effects of rumination. We hypothesized that effects of rumination were stronger for mid/post-pubertal adolescents than pre/early-pubertal adolescents, and tested this with pubertal stage x rumination interaction terms. To disentangle effects of age and puberty, we also tested age x rumination interactions.