Martijn van Teffelen

Interpretation bias modification for hostility 145 6 Several important limitations impact the present findings. First, the samples included participants with clinical levels of hostility that were both treatment- and non-treatment- seeking people. Although most people were screened in a treatment-seeking population (80%), treatment-seeking people made up (only) 38% of the final sample. It could be argued that people with clinical levels of hostility are more likely to decline participation, but Figure 1 shows that our sample composition is more likely due to treatment-seeking people that did not meet our inclusion criteria. However, this still implies that our results may not generalize to a treatment-seeking sample that for instance shows increased numbers of experienced psychosocial stressors, comorbidity, tendencies for interpersonal conflict and premature treatment discontinuation. As a related issue, the present sample included more men than women. Given that men express aggression more physically than women (Björkqvist et al., 1992; Genovese et al., 2017) this could have amplified the strength of the present findings. When we explored CBM-I effects on behavioral aggression for men only, its effect slightly increased from d = 0.28 using the complete sample to d = 0.41. A second limitation of the present study is that it did not include a follow-up measurement. We therefore do not know whether the results sustain over time. Third, there is currently no consensus in the CBM-I literature in terms of dose-response effects. We opted for eight 20-minute sessions, but we urge future studies to investigate the optimal dose-response effect. A fourth drawback of the study was that we omitted to define interpretation bias as a main outcome prior to the study, while we did base the a priori power analysis on interpretation bias only. Fifth, the present study contained and active and a control condition. Perhaps, the observed findings were impacted by the fact that people who score high on hostility do not like to be manipulated. However, participants were explicitly told they were going to be randomly allocated to an intervention study for hostility with active or control condition. We also asked the participants if they thought whether or not they thought the interventions was beneficial for them. Within the CBM-I and AC conditions, respectively 72.5% and 40.0% of participants indicated they found the intervention useful. However, the perceived usefulness per condition did not depend on baseline hostility levels. Last, we originally intended to exclude people who actively used alcohol or drugs. After additional scrutiny of the literature, however, we could not find convincing evidence that supported this criterion. On the contrary, literature showed that bias modification studies are conducted and shown to be efficacious in samples that are on active alcohol use (Wiers et al., 2015). That study showed for example that the alcohol approach bias significantly reduced, but non-differentially from active control training. In the present study, analysis showed that alcohol or drug consumption on the same day prior to CBM-I sessions did not impact the pattern of results. We recommend future studies on CBM-I to further disentangle the influence of alcohol and substance consumption on CBM-I efficacy.