Is there a Reliable Way of Measuring AB?
The finding that the internal consistency of the DPT was unacceptable in both parents and children is consistent with previous studies of adults [
13‐
21] and children [
19,
22]. Van Bockstaele, Notebaert, et al. found the VST to be a more reliable measure of AB in unselected adults [
24], whereas findings from child samples suggested poorer reliability [
19,
22]. Interestingly, in the current study we found the opposite pattern of findings: reliability of the behavioral index of the VST (AB
VST-RT) was acceptable in children but questionable to unacceptable in parents. The VST was theorized to have better internal consistency because of the longer exposure times in the task (increasing task: error variance). The finding that the VST had poor reliability in the current sample of adults may be explained by the fact that we included parents with a history of depression whereas the previous study of VST reliability recruited undergraduate students without elevated symptoms of psychopathology. It is important that the reliability of tasks is considered in relation to specific populations since mental disorders such as depression are known to influence reaction times. Similarly, the poorer reliability of the VST in previous studies of children may be explained by the fact these studies included younger children (aged 7–9 years) whereas children in the current study were older (9–14 years). Contrary to expectations that ET measures would be associated with superior reliability, the reliability of the ET index of the VST and the PVT orientation index also showed unacceptable reliability. The PVT maintenance index showed superior reliability: reliability in the child sample was questionable to good, and in the parent sample it was good to excellent. The fact that reliability was higher for the maintenance (versus orientation) index corresponds with other reports suggesting that the ET reliability is higher for indices measured over longer (versus shorter) periods of time [
25]. In summary, whilst the DPT showed poor reliability in our study, behavioral indices of the VST task along with the maintenance index of the PVT showed better reliability. Whereas the poor reliability associated with AB tasks may reflect methodological limitations of the tasks themselves, others have argued that AB itself may be an unreliable construct, varying within individuals within short periods of time [
95].
Do Children of Depressed Parents Show a more Negative AB?
In light of the poor reliability and construct validity (no correlations with depressive symptoms in children or parents) of the DPT in the current study, it is perhaps not surprising that the task failed to differentiate between HR and LR children. Nevertheless, our findings contradict those of many other studies using similar methodology and samples [
54‐
60] which is hard to explain. It is theoretically possible that the DPT showed superior reliability in previous studies compared to ours, however the psychometric properties of the task in these studies are not reported and given that numerous studies have reported poor reliability of the task in adult [
13‐
21] and child [
19,
22] samples, superior reliability is unlikely. Our null-findings are also unlikely to reflect a lack of power, since our sample size is comparable, if not larger, than most previous studies’. For example, one study that found daughters of depressed mothers to show more negative AB than daughters of never-disordered mothers included just 20 HR children [
54]. Other studies that had sample sizes comparable to ours (36 and 38 HR children [
55,
56]) found negative AB only in HR children who were female [
55] or had a parent with AB away from positive information [
56]. Two studies that investigated larger samples (241 and 244 HR children respectively [
57,
58]) did also not find significant main effects of group but more negative AB only in sub-groups of HR children that showed elevated cortisol reactivity to stress [
58] or were female [
57]. Unfortunately, our sample was too small to explore whether gender or direction of parental AB might explain our null-effects. Of course one possible explanation is that a publication bias exists in the field of AB in youth depression. Whilst it is common knowledge that small sample sizes reduce the chances of finding a true effect (Type II error), it is less well known that they also result in over-inflated effect sizes due to only large effects passing statistical thresholds [
96]. To this extent, it is possible that other studies in children of depressed parents with larger sample sizes have been conducted but failed to be published due to their lack of significant effects. A meta-analysis of AB in at-risk and depressed youth which estimates the likelihood of a publication bias in the field could inform this hypothesis and provide a valuable contribution to future research. In summary, based on the poor reliability of the DPT in our sample and the modest sample sizes of previous studies, we urge extreme caution in the interpretation of their findings.
The null-findings between HR and LR children from the AB
VST-ET index and the PVT orientation index can also plausibly be attributed to the poor psychometrics of the tasks in the current sample. However, the null-effects in relation to the AB
VST-RT index in HR versus LR children cannot since reliability of this index was acceptable. For the PVT maintenance index, reliability was questionable to good, yet there was no evidence of AB in HR versus LR children. It is worth mentioning that a main effect of group was observed for the maintenance index: HR children spent less time looking at the faces (regardless of emotion) than LR children. This may reflect an avoidance-based emotion regulation strategy which HR children have developed as a result of potentially inconsistent emotions expressed by their parent. Nevertheless, this finding was not predicted and its interpretation remains speculative. It is worth considering whether other methodological factors could explain the null-findings in HR and LR children in the AB
VST-RT and PVT maintenance indices. Firstly, the findings cannot be attributed to difficulties children had in completing the tasks since the accuracy rate for the VST was 94% and the PVT did not involve any active response. A second possibility is that children were not in enough of a negative mood state for AB to be observed (cognitive models propose that cognitive biases only emerge under conditions of stress or negative mood [
61,
62]). However, the current data suggest that the mood induction was successful in inducing a negative mood. Could the null findings in the HR versus LR children relate to sample characteristics? Since some of the HR children were recruited through a preventive intervention trial, it is theoretically possible that the intervention positively influenced AB in the HR group. However, given that just 10 of the children had participated in the intervention by the time they took part in the present study, this also seems an unlikely explanation for our findings. More plausible is perhaps that children whose parents had the motivation to sign up to an extensive intervention (see footnote 6) despite having a history of depression are less vulnerable to depression in the first place than children of depressed parents who do not sign up for such an intervention. Nevertheless, the fact that group differences (HR versus LR) have been observed for an implicit measure of IB in the same sample [
67], the current null findings are unlikely to be due to sample characteristics.
One additional explanation for the lack of differences in AB between HR and LR children might be that AB are rather correlates of depressive symptomatology that arise as a consequence of the disorder rather than antecedents that act as cognitive vulnerabilities or risk factors contributing to the development of the disorder. This might explain why our results are not in line with studies that also included youth with elevated levels of depression in their HR group [
52,
59,
60]. Once established, AB may exacerbate depressive symptoms and are likely to contribute to the maintenance of the disorder [
97]. To date, only few studies have investigated to what extent cognitive biases are risk factors for depression vs. consequences of depression. Studies on memory biases [
98] and IB [
99] have found negative biases to be present in both depressed as well as at-risk youth compared to low-risk youth, but to be more pronounced in depressed compared to at-risk youth. However, regarding AB we found in a subsequent study [
100] evidence for currently depressed youth to dwell longer on disorder-specific emotional information (i.e., sad faces) than healthy youth, particularly healthy youth at high risk for depression, suggesting that HR youth might even show AB in opposite directions to depressed youth.
Do Parents with a History of Depression Show a Negative AB?
Although one previous study failed to find evidence of increased attention to negative stimuli in remitted depressed adults [
101], the current findings contrast with the majority of previous studies which have found negative AB in adults with a current episode [
30] or past history [
102‐
105] of depression. This is perhaps unsurprising given that the tasks generally showed poor reliability in parents. However, the lack of effects in relation to the maintenance index of the PVT was unexpected since reliability of this task was good to excellent. The null-results might be explained by the fact that almost all parents in the HD group had previously received psychotherapy, mostly CBT, which is known to modify cognitive biases [
106] and might therefore have reduced their AB. This may be particularly true in situations where conscious processing and intrinsic guidance of attention is possible, as in the PVT. Another possibility might be that some parents were taking psychotropic medication, which is known to influence AB [
107]. We exploratorily compared AB scores of HD parents who were taking psychotropic medication with those of HD parents who were not. Since no differences emerged (all
ps > .1) this is unlikely to account for the null effects. It is unlikely that participation in the preventive intervention through which some participants were recruited accounted for the null-effects, since the intervention targeted parenting strategies but was not designed to modify parents’ symptoms of depression per se. Finally, the severity of parents’ depression (or anxiety) symptoms is unlikely to account for the null-effects, since there was no evidence that any of the AB indices correlated with BDI (depression) or STAI (anxiety) scores. Unfortunately the study was not sufficiently powered to investigate whether single-episode (
n = 8) versus recurrent (
n = 34) depression could explain the null effects. Nor whether comorbid disorders (
n = 12) versus no comorbidity (
n = 32) could explain the effects. Since the diagnostic status of the second parent was not systematically assessed in the study it was also not possible to explore whether the null-effects were due to having one versus two parents affected by depression.
Are Child and Parent AB Correlated?
A key hypothesis was that child and parent AB would be associated with another. However, there was no evidence of this for any of the measures. This finding was unexpected since a previous DPT study found AB in parents with a lifetime emotional disorder and their children to be related [
56]. However, there are numerous methodological differences between the latter study and our study which make comparisons difficult. For example, in the latter study the majority of parents had an anxiety disorder and only 14 had depression, thus it is possible the positive child-parent association was driven by parents with an anxiety disorder. Secondly, in the latter study both children and parents viewed adult facial stimuli whereas in the current study facial stimuli were matched for age (i.e., children viewed child stimuli). Thus it is possible that our null effects are due to AB being specific to the age of the models. Finally, the robustness of the findings from the study of children of parents with lifetime emotional disorder is questionable: a parent–child AB correlation was only found in the 38 HR children (but not the 29 LR children), and only between child negative AB and parent positive AB (negative correlation) but not parent negative AB. Given the relatively small number of HR children and the specificity of the findings to one form of parent AB, we urge caution in the interpretation of these effects. In fact, our null-effects are also in line with a study of AB in anxious parents and their offspring: There was no evidence for a correlation between parent AB and child AB [
19]. Finally, the current findings are also somewhat consistent with additional findings from the same sample in relation to measures of IB [
67]: Although HR children and their HD parents showed an implicit negative IB, there was no evidence of transgenerational correlation in this IB. Nevertheless, the fact that we found no evidence of AB in HD parents limits the conclusions that can be drawn from the current study about the transgenerational role of AB.
Strengths and Limitations of Study
A major strength of the current study is the inclusion of multiple measures of AB; both behavioral (DPT, VST) and ET-based (VST, PVT). Findings from previous studies of AB in children of depressed parents have been difficult to compare due to heterogeneity in their methodology. In this study, consistent null-results across three different tasks enable us to draw conclusions with more certainty and to directly compare the psychometric properties of the various measures within a single sample. A second strength is the use of standardized diagnostic instruments to categorize parental and child psychopathology. Evaluating parents’ psychiatric status via standardized clinical interviews is a more valid means of assessment than self-report [
108]. Standardized clinical interviews were administered not only to all participating children and parents but also to the second parent in the ND/LR families to ensure that neither of the child’s parents had a history of depression or any other psychiatric disorder. Finally, the study makes important contributions towards the open science movement (
https://www.cos.io/). Recommendations for addressing the replication crisis in psychology include pre-registration, full-disclosure of analysis methods and publication of non-significant findings [
109,
110]. This study was registered on the OSF prior to data collection. It is the first study of AB in children of depressed parents, and one of very few of youth in general [
19,
22], to report the psychometric properties of the experimental tasks employed. In contrast with most ET studies, we report data pre-processing steps and the data analysis strategy in great detail. Finally, in light of the “filedrawer” problem, and publication bias the publication of non-significant findings makes a valuable contribution to the existing literature.
Some limitations of the study are also important to mention. The relatively modest sample size means that we cannot exclude the possibility that our null-effects are due to a lack of statistical power. Although we based our sample size calculation on previous studies using similar methodology in the same population [
54,
55], in hindsight these effects are likely to have been over-estimates of the true effect. Unfortunately, we had no meta-analysis of AB effect sizes in youth depression to guide our sample size calculation. A second limitation of the sample is the fact that some parents in the HD group and their HR children were recruited via an ongoing preventive intervention for families affected by depression and may therefore be unrepresentative of families with a depressed parent in general. However, as mentioned above, this is unlikely to be enough of an explanation for our null-results. Finally, whilst ET methodology has many advantages over behavioral measures of AB, it does also carry some limitations [
25]. Firstly, the psychometric properties of ET measures are not always superior to RT measures and are under-investigated in many populations (e.g., in children and adolescents). Secondly, it must be acknowledged that it cannot detect changes in covert attention since covert attention can occur without eye movements. Nevertheless, covert attentional processes are largely thought to mediate overt attentional processes.
Clinical Implications
Given the lack of an association between parent and child AB, it is unlikely that AB plays a role in the transmission of depression risk from parent to child. In line with findings on the association between parent and child IB in the same sample [
67], it is possible that whilst reflective cognitive factors are passed on from parent to child, implicit processes are not. Our findings provide further evidence that the validity of apparently positive findings from previous DPT studies of AB in HR children and their parents be called into question [
54‐
60]. Since just one study has shown evidence of cross-sectional parent–child associations of AB and no study has yet examined the role of AB in the onset of depression prospectively, we are far from implicating AB in the transgenerational transmission of depression risk. A recent meta-analysis demonstrates a clear need for improved preventive interventions for the children of depressed parents [
111]. In the increasing popularity of cognitive bias modification of attention (CBM-A) paradigms, the current study suggests the field is not ready for CBM-A interventions for the children of depressed parents. Similarly, although the DPT has been used to assess changes in AB following therapeutic interventions, the poor psychometric properties generated in this study suggest this is an inappropriate task for such purposes.
Future Research
The focus of AB research in the field of depression has largely been on biases towards
negative information. However, recent meta-analyses demonstrate that depression is also characterized by an avoidance of
positive stimuli [
112]. It seems plausible that avoidance of positive stimuli may be involved in the transgenerational transmission of depression, and this may be a valuable area of future research. Based on the findings of the current study, other researchers are urged to evaluate and report the psychometric properties of the tasks they use to investigate AB. The poor psychometric properties of the DPT observed in this study combined with the lack of reporting on psychometric properties in previous studies of children of depressed parents suggests researchers should be cautious about using this paradigm in this sample. Note that one study found that the poor reliability of the DPT is in part influenced by issues to do with data preparation and analysis, which can be optimized to achieve more moderate reliability [
113]. However, as others note [
65], even when these recommendations are followed, reliability does not necessarily improve [
114]. As such, there is also a need for alternative measures of AB to be developed with improved psychometric properties that also adequately address the natural variability of the construct AB itself [
95]. Although ET measures were expected to show superior reliability to the DPT, this was not entirely the case, suggesting that one should not assume ET indices to be necessarily superior. A valuable line of future research would also be to develop more reliable ET measures of AB.
Of note, we observed some developmental differences in task reliability: the maintenance index of the PVT showed better reliability in parents than children whereas the behavioral VST index showed better reliability in children than parents. Future studies might involve adapting existing tasks to improve reliability within age groups. A related avenue of future research is to develop a clearer criteria as to what constitutes acceptable reliability for experimental tasks [
13]. Once AB can be reliably measured and observed in the children of depressed parents, an important next step would be not only to assess correlations between parents and children cross-sectionally [
56] but also longitudinally [
115]. Such studies are in a better position to determine the extent to which AB observed in HR children are responsible for the increased onset of depression.