8 333 GENERAL DISCUSSION There is always the possibility (i.e., β), however, that our non-significant findings represent false negatives (Type II error), or that small effects or differences exist but were undetectable due to our sample size. Acknowledging this, we recognise that our a-priori power might be limited to detecting medium-sized effects, and thus, we may have erroneously concluded a lack of difference or effect. Nevertheless, the consideration of practical relevance is crucial here. Small effects, as noted by J. Cohen (1988), are challenging to detect and might not translate into real-world, noticeable, improvements. A small effect size should therefore be considered negligible (Hojat & Xu, 2004). Even if larger sample sizes were employed, potentially revealing small effects, their practical importance might still be minimal. Despite this, however, our results do not provide insights into the strength of evidence supporting the null hypothesis (Rouder et al., 2009). Analytic approaches comparing the likelihood of findings under the null hypothesis to the likelihood under the alternative hypothesis are necessary. Bayesian analyses offer a solution by computing a likelihood ratio, indicating under which hypothesis – null or alternative – the observed finding is more likely to have occurred (Leppink et al., 2017; Rouder et al., 2009). This approach provides a spectrum of evidential strength, ranging from ‘not worth more than a bare mention’ to substantial, strong, very strong, and decisive evidence (Jeffreys, 1961). Thus, a comprehensive understanding of how strong the evidence for the current non-significant findings is, requires Bayesian hypothesis tests. Future research should consider incorporating these analyses to address the questions related to evidence favouring the null hypothesis. A second limitation of the current dissertation is that none of the studies collectively explored all hypothesised associations simultaneously or provided a comprehensive overview of these associations. In the general introduction of this dissertation, we posited that nonspecific factors might exert a direct impact on intervention outcomes, as well as an indirect effect through variables related to engagement. Additionally, we theorised that these nonspecific factors might interact with each other and with specific factors in predicting both engagement and intervention outcomes. While each study in this dissertation delved into specific facets or sub-relations of the hypothesised associations between specific and nonspecific factors, engagement-related variables, and mental health outcomes, none comprehensively examined all these associations concurrently within the same experimental setup and population. Future research should aim to address this gap by providing a holistic understanding of the interplay between these various factors. It is important to note, however, that Type I errors become more likely when conducting multiple tests on the
RkJQdWJsaXNoZXIy MTk4NDMw