Page 258 - Like me, or else... - Michelle Achterberg

Chapter 9
subtle brain-behavior associations, specifically in children. That is, using multiple smaller samples (n<30) in chapter 2 did not reveal the social rejection specific neural activity that was found using a larger sample (n>300) in chapter 4. Moreover, the independent sample for ROI selection (n=41) in chapter 5 had too little statistical power to reveal the whole brain-behavior associations that were reported with the exploratory analyses (n>300). An important methodological objective that follows from this thesis is that multiple samples are not necessarily better than large samples (or vice versa), but that they serve different purposes. Replicability is extremely important for confirming findings (Ioannidis, 2005; Schmidt, 2009), but for explorative discoveries we need a lot of statistical power and therefore large samples (Mumford and Nichols, 2008; Button et al., 2013). This is especially true for developmental neuroscientific studies, as the attrition rate in children often is high (O'Shaughnessy et al., 2008; Raschle et al., 2012; Fassbender et al., 2017a).
Control your head motion: Attrition biases
Children are more prone to head motion during the MRI scan than adolescents and adults (O'Shaughnessy et al., 2008; Raschle et al., 2012; Fassbender et al., 2017a). To limit the confounding effect of head motion on MRI findings, it is important to exclude participants that exceed a specific threshold of head motion (Power et al., 2015). This often results in an underrepresentation of children in cohort-sequential longitudinal studies, an issue that can be overcome by oversampling children during data acquisition. However, excluding participants who display excessive head motion might induce an additional bias: it is likely that participants who have difficulty regulating their head motion also experience difficulty regulating their emotions and behaviors. Indeed, studies showed a significant association between head motion and motor control (Zeng et al., 2014; Ekhtiari et al., 2019). This indicates that participants with the most behavioral control problems are the first to be excluded in MRI research (Kong et al., 2014). This bias is almost insurmountable, but must be kept in mind when interpreting neuroscientific studies on emotion regulation and behavioral control. More and more methods to deal with head motion during MRI scan acquisition are being developed, for example by using real-time monitoring of head motion (Dosenbach et al., 2017) or customized head molds (Power et al., 2019), which might enable future studies to exclude less participants and thereby minimize attrition bias.
fMRI: State of mind or state of mess?
The reliability of functional MRI, specifically experimental (task-based) fMRI has been heavily debated in recent years (Nord et al., 2017; Elliott et al., 2019b; Frohner et al., 2019). The variability observed in fMRI blood oxygen level dependent (BOLD) signal and the poor test-retest reliability in developing populations is a big concern for the field of developmental neuroscience (Herting
256