c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 Available online at www.sciencedirect.com ScienceDirect Journal homepage: www.elsevier.com/locate/cortex Research report A purely confirmatory replication study of structural brain-behavior correlations Wouter Boekel a,*, Eric-Jan Wagenmakers a, Luam Belay a, Josine Verhagen a, Scott Brown b and Birte U. Forstmann a a b University of Amsterdam, Amsterdam, The Netherlands University of Newcastle, Australia article info abstract Article history: A recent ‘crisis of confidence’ has emerged in the empirical sciences. Several studies have Received 9 April 2014 suggested that questionable research practices (QRPs) such as optional stopping and se- Reviewed 20 May 2014 lective publication may be relatively widespread. These QRPs can result in a high propor- Revised 13 October 2014 tion of false-positive findings, decreasing the reliability and replicability of research output. Accepted 17 November 2014 A potential solution is to register experiments prior to data acquisition and analysis. In this Action editor Chris Chambers study we attempted to replicate studies that relate brain structure to behavior and Published online xxx cognition. These structural brain-behavior (SBB) correlations occasionally receive much attention in science and in the media. Given the impact of these studies, it is important to Keywords: investigate their replicability. Here, we attempt to replicate five SBB correlation studies Preregistration comprising a total of 17 effects. To prevent the impact of QRPs we employed a preregis- Confirmatory tered, purely confirmatory replication approach. For all but one of the 17 findings under Replication scrutiny, confirmatory Bayesian hypothesis tests indicated evidence in favor of the null Brain-behavior correlations hypothesis ranging from anecdotal (Bayes factor < 3) to strong (Bayes factor > 10). In several studies, effect size estimates were substantially lower than in the original studies. To our knowledge, this is the first multi-study confirmatory replication of SBB correlations. With this study, we hope to encourage other researchers to undertake similar replication attempts. © 2015 Published by Elsevier Ltd. 1. Introduction In the last few years, the need for confirmatory replication studies has become increasingly evident. Recent studies have suggested that the empirical sciences are bedeviled by the use of questionable research practices (QRPs; John, Loewenstein, & Prelec, 2012; Simmons, Nelson, & Simonsohn, 2011). These practices include, for instance, optional stopping (i.e., continuing data collection until p < .05) and cherry-picking (e.g., reporting only those variables, conditions, or analyses that yield the desired result). In combination with the ubiquitous file drawer problem (Rosenthal, 1979), the use of these QPRs results in a high false-positive rate, such that many significant findings may in fact be false (Ioannidis, 2005). This * Corresponding author. Nieuwe Achtergracht 129, 1018 WS, Amsterdam. E-mail address: W.E.Boekel@uva.nl (W. Boekel). http://dx.doi.org/10.1016/j.cortex.2014.11.019 0010-9452/© 2015 Published by Elsevier Ltd. Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 2 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 realization has brought about a crisis of confidence in the replicability and reliability of published research findings (Ioannidis, 2012; MacArthur, 2012; Pashler & Wagenmakers, 2012). A recent study by Button, Ioannidis, Mokrysz, Nosek, Flint et al. (2013) showed that this crisis of confidence extends to the neurosciences. The crisis of confidence can be reduced in several ways. One powerful remedy is to eliminate QRPs by preregistering experiments prior to data acquisition and analysis, resembling the standard operating procedure mandated in the case of clinical trials (Chambers, 2013; De Groot, 1969; Goldacre, 2009; Wagenmakers, Wetzels, Borsboom, van der Maas, & Kievit, 2012; Wolfe, 2013). In this article we apply study preregistration to assess the replicability of a series of findings in cognitive neuroscience. Research in cognitive neuroscience aims to investigate the link between brain and behavior. Recently, researchers have exploited significant advances in anatomical magnetic resonance imaging (MRI) to detect subtle differences in brain structure associated with differences in behavioral measures (Kanai & Rees, 2011). For example, in a study that received much attention in science and the media, Kanai, Bahrami, Roylance, and Rees (2012) found that individuals with a relatively large grey matter (GM) volume in specific brain regions have more Facebook friends. Other studies have reported structural brain-behavior (SBB) correlations between properties of grey and/or white matter (WM) and behavioral measures such as choice reaction time (RT) (Tuch et al., 2005), control over speed and accuracy in decision making (Forstmann et al., 2010), percept duration in perceptual rivalry (Kanai, Bahrami, & Rees, 2010; Kanai, Carmel, Bahrami, & Rees, 2011), components of attention (i.e., executive control and alerting; Westlye, Grydeland, Walhovd, & Fjell, 2011), response inhibition (King et al., 2012), metacognitive ability (i.e., the ability to evaluate one's perceptual decisions; Fleming, Weil, Nagy, Dolan, & Rees, 2010), aspects of social cognition (i.e., social network size; Bickart, Wright, Dautoff, Dickerson, & Barrett, 2011; social influence; CampbellMeiklejohn et al., 2012), distractibility (Kanai, Dong, Bahrami, & Rees, 2011), political orientation (Kanai, Feilden, Firth, & Rees, 2011), sensitivity to reward and approach motivation (Xu et al., 2012), moral values (Lewis, Kanai, Bates, & Rees, 2012), and empathy (Banissy, Kanai, Walsh, & Rees, 2012). Motivated by the increase in number and prominence of SBB correlations, as well as the general uncertainty regarding the reliability of non-preregistered research findings, we attempted to replicate a subset of the above-mentioned studies in a purely confirmatory fashion. It should be noted that conceptual replications, wherein a hypothesis from the original study is tested in a different experimental paradigm, do not provide reliable evidence for or against the robustness of the respective finding. Instead, only direct replications, wherein all relevant aspects of the original study are repeated can support or oppose the original finding (Pashler & Harris, 2012). Here, we report a preregistered, purely confirmatory replication of a subset of five SBB correlation studies selected from recent literature based on the brevity of their behavioral data acquisition. The transparency conveyed by a confirmatory design helps to avoid common pitfalls in neuroscience (and other sciences) such as the use of nonindependent analysis (Vul, Harris, Winkielman, & Pashler, 2009), double dipping (Kriegeskorte, Simmons, Bellgowan, & Baker, 2009), obscure data collection and analysis techniques which increase false-positive rates (Simmons et al., 2011), confirmation and hindsight bias on the part of the researcher (i.e., the tendency to confirm instead of disconfirm one's beliefs and the tendency to judge events more predictable after they have occurred, respectively; Wagenmakers et al., 2012). A strictly confirmatory framework was ensured by publishing a 'Methods and Analyses document' (M&A; http:// confrepneurosci.blogspot.nl/2012/06/advanced-methodsand-analyses_26.html) online before any data were inspected or analyzed (as recommended by several researchers, e.g., Chambers, 2013; De Groot, 1969; Goldacre, 2009; Wagenmakers et al., 2012; Wolfe, 2013). This M&A document was sent to the corresponding authors of the original studies. All authors agreed to the replication attempt and the processing pipeline as outlined in the M&A document. Any analysis not outlined in the M&A document will be labeled 'exploratory' (as recommended by Wagenmakers, Wetzels, Borsboom, & van der Maas, 2011). We confined our hypotheses to the direction and location of the SBB correlations reported in the original articles. For instance, Kanai et al. (2012) reported a positive SBB correlation between GM density in left amygdala and the number of friends on Facebook; consequently the to-be-replicated hypothesis postulates a positive SBB correlation between the same variables in our sample. This order-restriction of the hypotheses has two benefits. First, it allowed us to use one-sided as opposed to two-sided hypothesis tests, which are more specific and statistically more powerful. Second, it allowed us to focus our analyses on specific regions in the brain, i.e., regions of interest (ROI), instead of searching the whole brain for SBB correlations. This way we circumvent the need for multiple comparisons corrections that are required in whole-brain analyses. In order to quantify the evidence that the data provide for and against the null-hypothesis, we opted for a Bayesian hypothesis test for correlations and computed Bayes factors (BF; Jeffreys, 1961) instead of p-values (for a discussion of problems with p-values, see Edwards, Lindman, & Savage, 1963; Wagenmakers, 2007). Note that in contrast to Bayes factors, p-values are unable to quantify support in favor of the null hypothesis; a non-significant p-value indicates no more than a “failure to reject the null hypothesis”. The replication attempts will be considered successful if the corresponding Bayes factor supports the hypothesized relationship. Accordingly, a Bayes factor that supports the null hypothesis suggests a failed replication. In addition to this preregistered analysis, exploratory analyses examine estimates of effect size. It is possible that the Bayes factor supports the null hypothesis, but the estimated effect size is nevertheless close to the original effect size. To address this concern, an additional exploratory Bayes factor analysis compares the null hypothesis to an alternative hypothesis that incorporates the knowledge obtained from the original study (cf. Verhagen & Wagenmakers, 2014). These exploratory analyses occasionally provide a more nuanced perspective on the extent to which SBB correlations can be replicated. Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 2. Materials and methods 2.1. General methods Prior to inspection of the data, a preregistration protocol was published online (http://confrepneurosci.blogspot.nl/2012/06/ advanced-methods-and-analyses_26.html). This ‘Methods and Analyses’ (M&A) document described all data acquisition and analysis steps. Below we summarize the subparts of this M&A document which are applicable to the results described in this article. 2.1.1. Participants 36 undergraduate psychology students (mean age ¼ 20.12, SD ¼ 1.73; 18 females) with normal or corrected-to-normal vision were recruited from the participant pool of a previous 43-participant MRI study. The MRI study was recently conducted by Forstmann and Wagenmakers' research group at the University of Amsterdam and featured extensive Diffusion Weighted Imaging (DWI) and T1-weighted imaging. Hence, the additional effort involved in replicating the five studies consisted primarily in having participants complete a battery of behavioral tests. The experiments were approved by the local ethics committee of the University of Amsterdam. Participants received a monetary compensation for their time and effort. 2.1.2. Study selection We aimed to perform replications of a series of recent studies reporting correlations between brain structure and behavior. A review by Kanai and Rees (2011) provided us with many topical SBB correlation findings. In addition, several other studies were selected from previous literature. Brevity of behavioral data acquisition was the main selection criterion, to ensure that we would be able to replicate many SBB correlations while minimizing total acquisition time. 2.1.3. Study exclusion Several studies, although selected and described in the M&A document, were omitted from the final analyses based on several reasons: Kanai, Feilden, et al. (2011) found an SBB correlation between political orientation and brain structure in young adults, using a simple 5-point self-report measure ranging from very liberal to very conservative. The data that we acquired to replicate this contained insufficient variability in this self-report measure, and thus we excluded this study (mean: 2.26, SD: .57, range: 1e3; Supplementary Fig. S1 shows scatterplots of these data). The other three studies (Bickart et al., 2011; Kanai et al., 2010, Kanai, Carmel, et al., 2011) were excluded from final replication based on problems with the ROI masks sent by the authors of the original papers (e.g., missing masks, or masks which did not match coordinates reported in the original papers). Five studies remained for the final replication attempt. 2.1.4. General procedure The time between MRI-scanning and behavioral testing ranged from 25 to 50 days. Prior to the behavioral test session, participants received an information brochure and 3 signed an informed consent form. Participants were tested in individual computer booths. All instructions were shown on the computer screen or printed on top of the questionnaires. Participants began by filling out the following questionnaires: BIS/BAS (Carver & White, 1994), social network index (Cohen, 1997), social network size questionnaire (Stileman & Bates, 2007), cognitive failures questionnaire (CFQ) (Broadbent, Cooper, Fitzgerald, & Parkes, 1982), political orientation questionnaire (Kanai, Feilden, et al., 2011), moral foundations questionnaire (Graham, Haidt, & Nosek, 2009), and the interpersonal reactivity index (Davis, 1980). After completing the questionnaires, participants continued with the computerized tasks: Bistable SFM task (Wallach & O'Connel, 1953), random dot motion (RDM) task (Britten, Shadlen, Newsome, & Movshon, 1992; Gold & Shadlen, 2007), and the attention network test (Fan, McCandliss, Sommer, Raz, & Posner, 2002). The order of both questionnaires and computer tasks was randomized across participants. The total duration of the test session was 1 h and 30 min. A subset of these tasks and questionnaires (i.e., the ones connected to the five studies that were included in the final replication attempt) were analyzed. 2.1.5. MRI data acquisition DWI and T1-weighted images were collected on a 3T Philips scanner using a 32- channel head coil. For each participant, four repetitions of a multi-slice spin echo (MS-SE), single shot DWI scan were obtained using the following parameters: TR ¼ 7545 msec, TE ¼ 86 msec, 60 transverse slices, 2 mm slice thickness, FOV: 224  224 mm2, voxel size 2 mm isotropic resolution. For each slice, 32 diffusion-weighted images (b ¼ 1000 sec/mm2) along 32 directions were acquired, along with one image without diffusion weighting (b0 image, where b ¼ 0). In addition, a T1-weighted anatomical scan was acquired (T1 turbo field echo, 220 transverse slices of 1 mm, with a resolution of 1 mm3, TR ¼ 8.2 msec, TE ¼ 3.7 msec). 2.1.6. ROI-based analysis Our purely confirmatory approach allowed us to circumvent the multiple comparison problems present in whole-brain analyses. We extracted measures of brain structure from ROIs provided to us by the authors of the original papers. These measures were then correlated to the respective behavioral measure. This approach would not have been possible if the authors of the original authors had not provided us with the ROI masks of their findings. We would like to thank these authors for their cooperation and openness. 2.1.7. DWI analyses All DWI data (pre-)processing and analyses were carried out using FMRIB's Software Library (FSL, version 4.0; www.fmrib. ox.ac.uk/fsl). Per participant, all four runs of DWI were concatenated and corrected for eddy currents. Affine registration was used to register each volume to a reference volume (Jenkinson & Smith, 2001). A single image without diffusion weighting (b0; b-value ¼ 0 sec/mm2) was extracted from the concatenated data and non-brain tissue was removed using FMRIB's Brain Extraction Tool (BET; Smith, 2002) to create a brain-mask which was used in subsequent analyses. Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 4 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 DTIFIT (Behrens et al., 2003) was applied to fit a tensor model at each voxel of the data (Smith, Jenkinson, Woolrich, & Beckmann, 2004). Tract-Based Spatial Statistics (TBSS) were performed using FSL's default TBSS pipeline (Smith et al., 2006; http://www.fmrib.ox.ac.uk/fsl/tbss/index.html). First, fractional anisotropy (FA) images were slightly eroded and end slices were zeroed in order to remove likely outliers from the diffusion tensor fitting. Second, all FA images were aligned to 1 mm standard space using non-linear registration to the FMRIB58_FA standard-space image. Affine registration was then used to align images into 1  1  1 mm MNI152 space, and a skeletonization procedure was subsequently applied to a mean FA image resulting from averaging all individual MNIaligned images. Subsequently, the mean skeletonized FA image was thresholded at FA > .2 in order to accurately represent white-matter tracts. Participants FA data were then projected onto the mean skeletonized FA image and concatenated. In addition to using FA images, we repeated this processing pipeline for mean diffusivity (MD) and parallel eigenvalue (l1) images using the tbss_non_FA function in order to generate skeletonized MD and l1 files. As opposed to using voxel-wise permutation tests for significance, our purely confirmatory approach allowed us to extract and average FA/MD/l1 from ROIs based on spatial maps provided by the original authors. For the TBSS procedure, the spatial maps provided by the original authors were registered to the mean FA template generated by our TBSS procedure. This was done to maximize the overlay between the spatial maps and our study-specific skeletonized FA template. In order to exclude the possibility that this registration step might impact the final hypothesis test, additional exploratory analyses were performed without registering the spatial maps to our FA template. These analyses are not reported here, as their results did not differ from our main analyses in terms of interpretation (i.e., Bayes factors were comparable). After extracting FA/MD/l1 signal from the ROIs, we then used one-sided Bayesian correlation tests (described below) to quantify evidence in favor of either the null hypothesis (H0) or the alternative hypothesis (H1). In our analyses, H1 represents the presence of either a positive or a negative correlation (depending on the predicted direction of the correlation), and the H0 represents the absence of the predicted correlation. 2.1.8. Probabilistic tractography Bayesian estimation of diffusion parameters obtained using sampling techniques (BedpostX) was applied to the preprocessed DWI data. BedpostX uses a dual fiber model which can account for crossing fibers. Estimation of tract strengths (for the replication attempt of Forstmann et al., 2010) was conducted using probabilistic tractography (Behrens et al., 2003). Five thousand tracts were sampled from each voxel in the seed mask (right pre-supplementary motor area; Pre-SMA) at a curvature threshold of .2. Next, the number of samples that reach the classification target mask (e.g., right striatum) was measured. In addition, contralateral exclusion masks were used to discard pathways crossing over to the contralateral hemisphere before traveling to the classification target mask. The number of voxels for which a minimum of 10 samples reached the classification mask was divided by the total number of voxels in the seed mask, resulting in a value that represents the proportion of the seed mask that was probabilistically connected to the classification mask. A similar procedure was applied in the opposite direction (where the seed and classification masks were switched). Tract strength was defined as the average of the two proportions that resulted from the seed-to-classification and classification-to-seed analyses. 2.1.9. Voxel-Based Morphometry Voxel-Based Morphometry (VBM) was performed using FSL's default VBM pipeline (Douaud et al., 2007; http://fsl.fmrib.ox. ac.uk/fsl/fslwiki/FSLVBM). First, non-brain tissue was removed from T1 images using BET. Second, brain-extracted images were segmented into GM, WM, and cerebrospinal fluid (CSF). GM images were non-linearly registered to GM ICBM-152, and averaged to create a study-specific template at 2 mm resolution in standard space. All GM images were then non-linearly registered to the study-specific template. During this stage, each voxel of each registered GM image is divided by the Jacobian of the warp field (Good et al., 2001). Images were smoothed using a Gaussian kernel with a sigma of 3 mm. As opposed to using voxel-wise permutation tests for significance, our purely confirmatory approach allowed us to extract and average GM volume from ROIs based on spatial maps provided by the original authors. We then used onesided Bayesian correlation tests (described below) to quantify evidence in favor of either H0 or H1. 2.1.10. Cortical thickness analysis Cortical reconstruction and volumetric segmentation was performed with the FreeSurfer image analysis suite (http:// surfer.nmr.mgh.harvard.edu/). The technical details of these procedures are described elsewhere (Dale, Fischl, & Sereno, 1999; Dale & Sereno, 1993; Fischl & Dale, 2000; Fischl, van der Kouwe, et al., 2004; Fischl, Liu, & Dale, 2001; Fischl, Salat, Busa, Albert, Dieterich et al., 2002; Fischl, Salat, et al., 2004; Fischl, Sereno, & Dale, 1999; Fischl, Sereno, Tootell, & Dale, 1999; Han et al., 2006; Jovicich et al., 2006; Reuter, Rosas, & Fischl, 2010; Reuter, Schmansky, Rosas, & Fischl, 2012;
gonne et al., 2004). FreeSurfer pre-processing included Se motion correction (Reuter et al., 2010) of volumetric T1weighted images, removal of non-brain tissue using a hybrid
gonne et al., watershed/surface deformation procedure (Se 2004), automated Talairach transformation, segmentation of the subcortical WM and deep GM volumetric structures (including hippocampus, amygdala, caudate, putamen, and ventricles; Fischl et al., 2002; Fischl, Salat, et al., 2004) intensity normalization (Sled, Zijdenbos, & Evans, 1998), tessellation of the gray/white matter boundary, automated topology correc
gonne, Pacheco, & Fischl, 2007), and tion (Fischl et al., 2001; Se surface deformation following intensity gradients to optimally place the gray/white and gray/CSF borders at the location where the greatest shift in intensity defines the transition to the other tissue class (Dale & Sereno, 1993; Dale et al., 1999; Fischl & Dale, 2000). Reconstruction of the GM/WM boundary and pail surface was manually checked for inaccuracies. Subsequently, ROI-labels were mapped onto individual brains and average cortical thickness (Fischl & Dale, 2000) was extracted per ROI, per participant. Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 5 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 2.1.11. General outlier rejection criterion In the M&A document that we published online prior to inspection of the data, we specified a general outlier rejection criterion. Any deviation of more than 2.5 standard deviations (SDs) from the respective mean results in an exclusion of the participant from the replication in which it is classified as an outlier (as such, a participant can still be included in a different replication, for which he or she was not classified as an outlier). 2.1.12. Confirmatory Bayesian hypothesis test for correlations Our main analysis goal was to grade the decisiveness of the evidence that the data provide for and against the presence of a correlation between the structural brain measures and the behavioral measures. This goal can be achieved by computing Bayes factors (Dienes, 2008; Jeffreys, 1961; Kass & Raftery, 1995; Lee & Wagenmakers, 2013; Rouder, Morey, Speckman, & Province, 2012; Rouder, Speckman, Sun, Morey, & Iverson, 2009). The Bayes factor compares the adequacy of two models; in our case, the first model is the null hypothesis H0 that postulates the absence of a correlation between the structural brain measures and the behavioral measures. The second model is the alternative hypothesis H1 that postulates the presence of a positive (or negative) correlation between the two measures. The Bayes factor quantifies the odds that the observed data occurred under H0 versus H1. For example, a Bayes factor equal to 5.2 indicates that the observed data are 5.2 times as likely to occur under H0 than under H1. In this way the Bayes factor provides a continuous measure of evidential support, and its interpretation does not require recourse to actions, decisions, or criteria of acceptance. To compute the Bayes factor for the Pearson correlation coefficient, we need to specify both H0 and H1. Jeffreys (1961) proposed a default test by assigning uninformative priors to the nuisance parameters (i.e., parameters common to H0 and H1) and a uniform prior distribution from 1 to 1 to the correlation coefficient r that is unique for H1 (Jeffreys, 1961, p. 291). Consequently, under Jeffreys' alternative hypothesis H1, each value of the correlation coefficient r is a priori equally likely. Inspired by Jeffreys' test we grade the decisiveness of the evidence by computing BF10, that is, the probability of the observed data under H1 versus H0: BF10 ¼ Z1 0 ð1 ð1 1 ðn r2 Þ2 n rrÞ 1Þ 3 2 dr (1) The number of data pairs is denoted by n, and r is the sample Pearson correlation coefficient. As indicated by the range of integration in Equation (1), we have adjusted Jeffreys test such that the alternative hypothesis is one-sided. The one-sided nature of this test is appropriate, since we intend to replicate SBB correlations, thereby committing to specific directions (as reported in the original studies). In Equation (1), the integration is from 0 to 1 implying a test for a positive correlation. In case of a test for a negative correlation we simply multiply one of the observed variables with 1. An R function to compute the BF in the above-mentioned way is freely available at http://www.josineverhagen.com/? page_id¼76. The evidential support that the BF01 gives to the null hypothesis can be categorized based on a set of labels proposed by Jeffreys (1961). Table 1 shows this evidence categorization for the BF01, edited by and taken from Wetzels and Wagenmakers (2012; Table 1, p. 1060). In short, a BF01 greater than 1 indicates that the data are more likely to occur under H0 than under H1. Equivalently, a BF01 lower than 1 indicates that the data are more likely to occur under H1 than under H0. The evidence categories apply to the BF10 (¼1/BF01; reciprocal of the BF01) in a reversed manner; e.g., a BF10 with a value between 10 and 30 provides strong evidence for H1 and a BF10 with a value between 1/10 and 1/30 provides strong evidence for H0. Thus, when we analyze data and find that, for instance, BF01 ¼ 6.5, this means that the data are 6.5 times more likely to have occurred under H0 than under H1; similarly, BF01 ¼ .2 means that the data are 5 times more likely to have occurred under H1 than under H0. The labels shown in Table 1 are useful because they facilitate scientific communication; nevertheless, the labels should not be over-interpreted. Many researchers may find the meaning of BF01 ¼ 6.5 clear without the help of the labels from Table 1. 2.1.13. Posterior probability distributions The posterior distribution is formed by combining the information or beliefs about the correlation available prior to the experiment (as expressed in the prior distribution), with the correlation observed in the data. In a situation where nothing is known about the correlation prior to the experiment, an uninformative uniform prior distribution can be used, in which every correlation between 1 and 1 has equal probability (Fig. 1 black line). In this situation, once a correlation has been observed, the posterior distribution will have a higher probability around the observed correlation and less probability at values further away (Fig. 1 red line). The posterior distribution represents the knowledge we have about the correlation of interest after observing the data. When we want to update this knowledge with a new experiment, the posterior from the previous experiment can be taken as the prior for the next experiment. This indicates that the correlation in the new study is expected to be similar to the correlation in the previous study, as the prior gives more probability to values closer to the previously observed correlation. When this informative prior distribution is Table 1 e Categories for the BF01. Bayes factor BF01 30 10 3 1 1/3 1/10 1/30 1/100 > e e e e 1 e e e e < Interpretation 100 100 30 10 3 1 1/3 1/10 1/30 1/100 Extreme evidence for H0 Very Strong evidence for H0 Strong evidence for H0 Moderate evidence for H0 Anecdotal evidence for H0 No evidence Anecdotal evidence for H1 Moderate evidence for H1 Strong evidence for H1 Very Strong evidence for H1 Extreme evidence for H1 Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 6 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 posterior distribution and a summary in terms of 95% credible intervals. Finally, for frequentist readers we provide p-values. Once again, these are labeled as exploratory given that we did not preregister the use of frequentist statistics in our M&A document. 2.2. Fig. 1 e Posteriors plot. Example of a posterior plot, showing uniform prior distribution (black line), the posterior after the original effect (red line), and the posterior after the replication effect (blue line), using the posterior as a prior distribution. updated by the correlation observed in a new experiment, the final posterior distribution will be identical to the posterior distribution had all data been analyzed together from the start (Fig. 1 blue line). We will also use the posterior distribution of the previous study in a different way, for model comparison. In this case, the posterior distribution from the original study is used to represent the hypothesis that the observed correlation is similar to the previous correlation. 2.1.14. Additional exploratory analyses In addition to the Bayesian test described above, we computed an additional Bayesian test in which H1 is specified not only to the direction of the effect found in the original study, but also to its effect size (Verhagen & Wagenmakers, 2014). In this way, this test answers the question ‘Is the effect from the replication attempt comparable to what was found before, or is it absent?’, whereas the original Bayesian test answers the question ‘Is the effect present or absent in the data from the replication attempt?’. We label this additional analysis exploratory as it was not described and published in the M&A document prior to inspection of the data. The replication Bayes factor compares evidence in favor of the null hypothesis of no effect, H0: r ¼ 0, with the evidence in favor of the alternative hypothesis that the effect is equal to the effect found in the original study, Hr: r ~ posterior distribution from original study. The resulting Bayes factor is similar to the Bayes factor in Equation (1), with the only difference that the replication Bayes factor is obtained by integrating over the posterior distribution from the first study instead of a uniform distribution. A more detailed description of the replication Bayes factor can be found in Appendix A. R code to perform this analysis can be found in this link http://www. josineverhagen.com/?page_id¼76. In addition to the Bayes factor tests, an intuitive assessment of the extent to which our results replicate the original studies can also be obtained by comparing the posterior distributions for the correlation coefficients in the original and replication studies. We facilitate such a comparison by plotting, for each of the five replication attempts, both the entire Study-specific methods Below we describe study-specific methods for the five experiments included in the final replication attempt remaining after study exclusion. For each experiment we describe the stimuli and procedure, behavioral analyses, structural brain analyses, and statistical tests based on hypotheses generated by the original papers. 2.2.1. Replication 1: Forstmann et al. (2010) 2.2.1.1. RDM TASK AND PROCEDURE. We used the same RDM task (Gold & Shadlen, 2007) as Forstmann et al. (2010). The task contained 360 trials in total, with 180 speed and 180 accuracy trials. The RDM cloud consisted of 60 coherently moving white dots and 60 randomly moving white dots, presented against a black background (see http://wouterboekel.com/CONFREP/ dots_loop.gif). A single dot consisted of 3 pixels and the entire cloud spanned 250 pixels. At the start of each trial, either a speed cue or an accuracy cue was presented for 1000 msec. The speed cue instructed participants to respond as quickly as possible. The accuracy cue instructed participants to respond as accurate as possible. The cue was followed by a fixation cross presented at the center of the screen for 500 msec. Subsequently, the RDM stimulus was presented for 1500 msec or until a response was made. Responses outside of this time window were ignored. Participants responded on a keyboard by pressing ‘a’ with their left index finger when they perceived a leftward motion and ‘l’ with their right index finger when they perceived a rightward motion. Immediately after the response, participants received a feedback message for 400 msec. On speed trials, the feedback read either ‘te traag’ or ‘op tijd’ (i.e., Dutch for ’too slow’ and ‘in time’). On accuracy trials, the feedback read either ‘fout’ or ‘goed’ (i.e., Dutch for ‘incorrect’ and ’correct’). 45-sec breaks were inserted after 120 and after 240 trials. The entire task lasted for approximately 20 min. 2.2.1.2. LBA MODEL. The linear ballistic accumulator (LBA; Brown & Heathcote, 2008) model decomposes the response time and accuracy measures into latent psychological processes. It assumes that when given a choice between two alternatives, evidence accumulates from a start point (A), at a certain speed (drift rate v), for both alternatives separately. When one of these accumulators reaches its response threshold (b), a decision is made in favor of the associated alternative. Response time is determined by the time taken to reach the threshold, plus an offset time for stimulus encoding and motor processes (non-decision time t0) (Fig. 2). The element of central interest here is response caution, which can be quantified via the threshold height in the LBA. We applied the same parameter constraints as Forstmann et al. (2010). In this design only one parameterdresponse threshold bdis free to vary with the speed vs accuracy cue, while all other parameters (width of start point distribution A, Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 7 make a note of the number of friends they have on ‘Facebook’ or an alternative comparable social network site such as ‘myspace’ or the Dutch ‘Hyves’ and bring it to the test session. The administration time is approximately 10 min. 2.2.2.2. BEHAVIORAL Fig. 2 e Schematic representation of the LBA model used in the replication of Forstmann et al. (2010). In the LBA model, the decision to respond either left or right is modeled as a race between 2 accumulators. Activation in each accumulator begins at a random point between zero and start point A and increases with time. The rate of increase is random from trial to trial, but is (on average) faster for the accumulator whose associated response matches the stimulus. A response is given by whichever accumulator first reaches the threshold b, and the predicted response time depends on the time taken to reach that threshold. drift rate v, variability of the drift rate s, and nondecision time t0) are fixed. Response caution is measured by subtracting start point A from response threshold b. 2.2.1.3. BEHAVIORAL DATA ANALYSIS. The behavioral measure of interest is the LBA flexibility parameter, assessing efficacy of changing response caution. It is assumed that “changes in response caution originate from adjustments of response thresholds (Forstmann et al., 2010; page 1516)”. Therefore, LBA flexibility was computed as the difference between the LBA caution estimates for the accuracy and the speed conditions. We fit the LBA model to each participants accuracy and RT distributions on speed and accuracy trials separately. The only parameter allowed to vary was the response threshold b. The resulting individual LBA flexibility estimates were imported into R software (R Foundation for Statistical Computing, http://www.R-project.org) for the Bayesian correlation test. 2.2.1.4. PROBABILISTIC TRACTOGRAPHY. We limited our tractography to delineate tracts that the authors found to correlate significantly with LBA flexibility. Hence, probabilistic tractography was performed only between right pre-SMA and right striatum. Here we used the same MNI-space masks for right pre-SMA and right striatum as were used in Forstmann et al. (2010). We performed the probabilistic tractography in accordance with the protocol stated in the general methods section (see above). Resulting tract strength values were corrected for age and gender using partial correlations, and were subsequently imported into R software for the Bayesian correlation test. Specifically, we tested for a positive correlation between right pre-SMAeright striatum tract strength and LBA flexibility. 2.2.2. Replication 2: Kanai et al. (2012) 2.2.2.1. SOCIAL NETWORK SIZE QUESTIONNAIRE AND PROCEDURE. Participants completed a Dutch version of the Social Network Size questionnaire (Stileman & Bates, 2007). This questionnaire consists of 9 items. One of its items is: “How many friends do you have on ‘Facebook’?”. We asked participants to DATA ANALYSIS. The behavioral measures of interest are online social network size (i.e., FBN) and real-world social network size. As was done in Kanai et al. (2012), answers to the 9 subquestions contained in this questionnaire were square-root transformed to correct for skewness. We computed the FBN as the square root of participants answer to the question: “How many friends do you have on ‘Facebook’?”. A normalized real-world social network size score (SNS) was computed per participant by averaging the z-scores for the questionnaire items 1, 2, 4, 5, 6, 8, and 9 after skewness correction. For each participant an online social network size (i.e., FBN) score and a real-world social network size (i.e., SNS) score was imported into R software for the Bayesian correlation test. 2.2.2.3. ROI GENERATION. Kanai et al. (2012) reported significant positive correlations between online social network size and GM volume within left middle temporal gyrus (MTG), right superior temporal sulcus (STS), right entorhinal cortex (EC), and bilateral amygdala. In addition, real-world social network size was positively correlated with GM volume only within right amygdala. We defined all these regions as our ROIs. Dr. Kanai kindly provided us with the spatial maps of these regions. 2.2.2.4. CORRELATIONAL ANALYSIS. For every participant, we extracted GM volume values from all voxels contained in the ROIs and averaged them. These GM volume measures were then corrected for age, gender and total GM volume. The corrected mean GM volume measures were imported into R software for the Bayesian correlation test. Specifically, we tested for positive correlations between FBN and mean GM volume within left MTG, right STS, right EC, and bilateral amygdala. Furthermore, we tested for a positive correlation between SNS and mean GM volume within right amygdala. 2.2.3. Replication 3: Xu et al. (2012) 2.2.3.1. BIS/BAS QUESTIONNAIRE AND PROCEDURE. Participants completed a Dutch version of the Behavioral Inhibition System/Behavioral Activation System scale (BIS/BAS; Carver et al., 1994). The BIS/BAS is a 20-item questionnaire. Our interest was focused on the BAS scale, which comprises 13 items (BASTotal) and has three sub-scales: Drive (BAS-Drive), Fun-Seeking (BAS-Fun), and Reward-Responsiveness (BAS-Reward). 2.2.3.2. BEHAVIORAL ANALYSIS. The behavioral measures of interest were BAS-Total scores and BAS-Fun scores. BAS-Total scores assess the sensitivity to signals of reward and nonpunishment. BAS-Fun scores assess the tendency to seek out new potentially rewarding experiences. For each participant these scores were imported into R software for the Bayesian correlation test. 2.2.3.3. ROI GENERATION. Xu et al. (2012) reported significant positive correlations between the BAS-Total scores and l1 within left corona radiata (CR) and left superior longitudinal fasciculus (SLF). Furthermore, they reported positive Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 8 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 correlations between the BAS-Fun scores and l1 as well as FA within left CR and left SLF. The authors also reported significant positive correlations between the BAS-Fun scores and MD within left inferior longitudinal fasciculus (ILF) and left inferior fronto-occipital fasciculus (IFOF). We defined all these WM tracts as our ROIs. Dr. Xu kindly provided us with the spatial maps of these areas. 2.2.3.4. CORRELATIONAL ANALYSIS. For every participant, we extracted FA, MD, and l1 values from all voxels contained in the respective ROIs and averaged them. These values were then corrected for age and gender using partial correlations. Unlike Xu et al. (2012), we did not need to correct for differences in education because our participants were all first-year Psychology students. The corrected mean WM tract measures per ROI were imported into R software for the Bayesian correlation test. Specifically, we tested for positive correlations between BAS-Total scores and mean l1 within left CR and left SLF. Furthermore, we tested for positive correlations between BAS-Fun scores and mean l1 as well as mean FA within left CR and left SLF. Finally, we tested for positive correlations between BAS-Fun scores and mean MD within left ILF and left IFOF. 2.2.4. Replication 4: Kanai, Dong, et al., (2011) 2.2.4.1. COGNITIVE FAILURES QUESTIONNAIRE AND PROCEDURE. Participants completed a Dutch version of the CFQ (Broadbent et al., 1982). 2.2.4.2. BEHAVIORAL DATA ANALYSIS. The behavioral measure of interest is distractibility as assessed by the CFQ. As in Kanai, Dong, et al., (2011), we quantified distractibility by computing the standard loadings derived from a previous factor analysis (Wallace, Kass, & Stanny, 2002). Specifically, we used the following 9 items: 1, 2, 3, 4, 15, 19, 21, 22, and 25. Scores on these items were imported into R software for the Bayesian correlation test. 2.2.4.3. ROI GENERATION. Kanai, Dong, et al. (2011) reported a significant positive correlation between CFQ scores and GM volume within left superior parietal lobe (SPL). Furthermore, the authors reported a negative correlation between CFQ scores and GM volume within left middle prefrontal cortex (mPFC). We defined these regions as our ROIs. Dr. Kanai kindly provided us with the spatial maps of these regions. 2.2.4.4. CORRELATIONAL ANALYSIS. For every participant, we extracted GM volume values from all voxels contained in the respective ROIs and averaged them. These GM volume values were then corrected for age, gender and total GM volume using partial correlations. The corrected mean GM volume values were imported into R software for the Bayesian correlation test. Specifically, we tested for a positive correlation between CFQ scores and mean GM volumes within left SPL, and for a negative correlation between these measures within left mPFC. Fan's website www.sacklerinstitute.org/users/jin.fan). The task included 2 runs of 96 trials and 20 practice trials. Each trial began with the presentation of a fixation cross in the center of the screen for variable durations (400, 800, 1200, or 1600 msec). Subsequently, one of three cues was presented for 100 msec: (1) no cue, (2) center cue (*, replacing fixation cross), or (3) spatial cue (*, above or below fixation cross). This was followed by the presentation of the target for a maximum duration of 1700 msec, or until a response was made. The target was an arrow in the center of a row of 5 arrows, presented either below or above the fixation cross. The flanking arrows consisted of either (1) two congruent arrows (pointing in the same direction as the target), (2) two incongruent arrows (pointing in the opposite direction of the target), or (3) two lines on each side of the target (neutral). Participants were instructed to report the direction (left or right) of the target arrow by pressing the spatially compatible key (‘left mouse button’ or ‘right mouse button’) with their left or right thumb. The entire experiment took approximately 15 min. 2.2.5.2. BEHAVIORAL DATA ANALYSIS. The behavioral measures of interest are executive control and alerting network scores, assessing the executive control and the alerting components of attention, respectively. We applied the same processing steps as described by Westlye et al. (2011) prior to computing these scores: “To remove outliers, all RTs >1500 msec and <200 msec were removed (...). Next, since error responses are assumed to originate from a different RT distribution than correct responses, we only analyzed correct responses. Also, because responses following erroneous responses typically are slower than responses following correct responses (posterror slowing), we also removed responses following erroneous responses. Since RTs are not normally distributed, we used median RT per condition as raw scores for each subject. (...). (page 348).” However, we did not adjust the component scores with the baseline RT in order to control for an effect of age on RT, because our participants form a homogenous age group (Psychology freshmen). Based on median RT, the executive control and alerting scores will be computed as follows:  Executive control ¼ RTincongruent Alerting ¼ ½RTno cue RTcongruent  RTcongruent RTcenter cue Š=RTcenter cue For each participant, the resulting scores were imported into R software for the Bayesian correlation test. 2.2.5.3. ROI GENERATION. For their subsample of young participants, Westlye et al. (2011) reported significant negative correlations between executive control scores and CT within left caudal anterior cingulate cortex (ACC), left superior temporal gyrus (STG), and right middle temporal gyrus (MTG). The alerting scores showed a significant negative correlation with CT within left superior parietal lobe (SPL). We defined all these regions as our ROIs. Dr. Westlye kindly provided us with the FreeSurfer labels of these areas. 2.2.5.4. CORRELATIONAL 2.2.5. Replication 5: Westlye et al. (2011) 2.2.5.1. ATTENTION NETWORK TEST. We used the same Attention Network Test as Westlye et al. (2011; downloaded from Dr. Jin ANALYSIS. For every participant, we extracted CT values from all voxels contained in the ROIs and averaged them. These CT measures were then corrected for age and gender using partial correlations. The corrected mean Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 9 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 3. Right Pre-SMA - Right Striatum Tract Strength CT measures were imported into R software for the Bayesian correlation test. Specifically, we tested for negative correlations between executive control scores and mean CT within left caudal ACC, left STG and right MTG. Furthermore, we also tested for a negative correlation between alerting scores and mean CT within left SPL. Results Below we describe study-specific results for the five experiments included in the final replication attempt remaining after study exclusion, comprising a total of 17 predicted effects. For each study, we briefly re-iterate the original findings, followed by our predictions based on these findings. We describe potential outlier exclusion and list the Bayes factors in favor of the null hypothesis (BF01). Furthermore, we describe the outcome of the additional exploratory Bayes factor analysis that uses an informative prior distribution (cf. Verhagen & Wagenmakers, 2014). 3.1. r = .03 BF01 = 3.90 .0 .5 1.0 1.5 LBA Caution Parameter (Accuracy - Speed) Fig. 3 e Scatterplot of replication 1: Forstmann et al. (2010). The relationship between LBA caution parameter (quantified by taking the difference in response caution between the accuracy and speed condition) and tract strength between right Pre-SMA and right Striatum, quantified by probabilistic tractography. Replication 1: Forstmann et al. (2010) Forstmann et al. (2010) reported that individual differences in tract strength from right pre-SMA to right striatum predict individual differences in control over speed and accuracy in a perceptual decision making paradigm. The original authors replicated their effect in an independent data set. In line with the original authors' theorizing and results, we hypothesized the presence of a positive correlation between pre-SMAstriatum tract strength and LBA flexibility. Three participants did not complete the behavioral task and were thus excluded from further analysis. Tract strengths of 2 out of 33 participants deviated more than 2.5 SDs from the group mean, and were thus excluded from this replication attempt. After outlier rejection, tract strength data ranged from .682 to .914, with a mean of .819 and a standard deviation of .061. LBA flexibility ranged from .020 to 1.554, with a mean of .578 and a standard deviation of .410. A one-sided Bayesian hypothesis test for positive correlations was performed on these data. Its result is shown in Table 2 and Fig. 3. The Bayes factor shows that there is moderate support for the null hypothesis of no correlation. In order to provide a complete report of the SBB correlation found here in comparison with the original finding, Fig. S2 shows posterior probability plots of this effect. The additional exploratory Bayes factor analysis with informative priors (Verhagen & Wagenmakers, 2014) shows that the data are extremely likely to have occurred under the null hypothesis compared to the proponent's hypothesis. Fig. S2 (bottom) shows posteriors for this exploratory Bayes factor analysis. The p-value indicated a failed replication. 3.2. Replication 2: Kanai et al. (2012) Kanai et al. (2012) showed that individual differences in the number of Facebook friends (FBN) and real-world social network size (SNS) are positively correlated with GM volume in several brain areas. The original authors replicated their effects in an independent data set. In line with the original authors' theorizing and results, we hypothesized positive correlations between FBN and GM volume in left MTG, right STS, right EC, and bilateral amygdala. In addition, we hypothesized a positive correlation between SNS and GM volume in right amygdala. One participant did not complete the FBN and two participants did not complete the SNS questionnaire, and were thus excluded from further analysis. One participant was excluded in 4 out of 6 Bayesian correlations, due to a GM volume measure deviating more than 2.5 SDs from the group mean. After outlier rejection, the following summary statistics describe our data: FBN: range: 10.0499e24.7386, mean: 17.096, sd: 3.788. SNS: range: 1.05 to .44, mean: .650, sd: .153. GM in left MTG: range: .411e.562, mean: .476, sd: .035. GM in right STS: range: .336e.595, mean: .484, sd: .062. GM in right EC: range: .521e.785, mean: .628, sd: .063. GM in left Amygdala: range: .636e.770, mean: .707, sd: .033. GM in right Amygdala: range .603e.772, mean: .670, sd: .035. One-sided Bayesian hypothesis tests for positive correlations were performed on these data. Results are shown in Table 3 and Fig. 4. In 5 out of 6 cases we Table 2 e Results of the one-sided Bayesian hypothesis test for a positive correlation. Data pair ROI Confirmatory norig Tract strength and LBA flexibility Pre-SMA to striatum 9 Exploratory nrep rorig rrep BF01 Evidence cat. BF0r Evidence cat. p-value 31 .93 .03 3.90 Moderate (H0) 180.20 Extreme (H0) .431 Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 10 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 Table 3 e Results of the one-sided Bayesian hypothesis tests for positive correlations. Data pair ROI FBN and GM volume left MTG right STS right EC left amygdala right amygdala SNS and GM volume right amygdala Confirmatory norig nrep rorig 125 125 125 125 125 34 35 35 34 34 65 33 Exploratory rrep BF01 Evidence cat. BF0r Evidence cat. p-value .35 .35 .35 .30 .32 .18 .11 .06 .14 .02 1.73 2.66 3.51 7.76 4.35 Anecdotal (H0) Anecdotal (H0) Moderate (H0) Moderate (H0) Moderate (H0) 1.06 2.06 3.32 9.56 3.88 Anecdotal (H0) Anecdotal (H0) Moderate (H0) Moderate (H0) Moderate (H0) .158 .261 .360 .779 .462 .26 .30 .57 Anecdotal (H1) .27 Moderate (Hr) .041 Fig. 4 e Scatterplots of replication 2: Kanai et al. (2012). (AeE) The relationship between the number of Facebook friends and GM in (A) left MTG, (B) right STS, (C) right EC, (D) left amygdala, (E) right amygdala. (F) the relationship between real world social network size and GM in the right amygdala. Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 11 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 find support for the null hypothesis. The Bayes factors show that there is moderate support for the null hypothesis in 3 out of 6 effects (i.e., no correlations between FBN and GM volume in right EC, and bilateral amygdala). Our data are ambiguous with regard to the correlations between FBN and GM volume in left MTG and right STS. In order to provide a complete report of the SBB correlations found here in comparison with the original findings, Figs. S3e8 show posterior probability plots of these effects. The additional exploratory Bayes factor analyses with informative priors (Verhagen & Wagenmakers, 2014) show that for two effects there is anecdotal evidence in favor of the null hypothesis compared to the proponent's hypothesis. For three effects there is moderate evidence in favor of H0, and for one effect there is moderate evidence in favor of Hr, compared to H0. Figs. S3e8 (bottom) show posteriors for these exploratory Bayes factor analyses. p-values indicate failed replications for 5 out of 6 effects. For the correlation between SNS and GM volume in right amygdala, the p-value indicates a successful replication. Table 4 and Fig. 5. In all cases we find support for the null hypothesis. The Bayes factors show that there is moderate or strong support for the null hypothesis in 3 out of 4 tests (i.e., no correlation between BAS-Total and l1 in left CR and SLF, no correlation between BAS-FUN and FA in left CR and SLF, and no correlation between Bas-FUN and l1 in left CR and SLF). Our data are ambiguous with regard to the correlation between bas-FUN and MD in left ILF and IFOF. In order to provide a complete report of the SBB correlations found here in comparison with the original findings, Figs. S9e12 show posterior probability plots of hese effects. The additional exploratory Bayes factor analyses with informative priors (Verhagen & Wagenmakers, 2014) show that for three effects there is extreme evidence in favor of the null hypothesis compared to the proponent's hypothesis, and for one effect there is moderate evidence in favor of H0. Figs. S9e12 (bottom) show posteriors for these exploratory Bayes factor analyses. All p-values indicate failed replications. 3.4. 3.3. Replication 4: Kanai, Dong, et al., (2011) Replication 3: Xu et al. (2012) Xu et al. (2012) reported that individual differences in diffusion measures of several WM pathways are positively correlated with individual differences in the tendency to seek out new potentially rewarding experiences (i.e., BAS-Fun) and the sensitivity to signals of reward and non-punishment (BASTotal). In line with the original authors' theorizing and results, we hypothesized a positive correlation between the BAS-Total scores and l1 within left CR and left SLF, a positive correlation between BAS-Fun and FA in left CR and SLF, a positive correlation between BAS-FUN and l1 in left CR and SLF, and a positive correlation between BAS-Fun and MD in left ILF and IFOF. One participant was excluded from l1 analyses due to WM structural measures deviating more than 2.5 SDs from the group mean. After outlier rejection, the following summary statistics describe our data: BAS-Total: range: 14e31, mean: 22.833, sd: 3.783. BAS-FUN: range: 5e12, mean: 7.667, sd: 1.821. FA in left CR and SLF: range: .649e.810, mean: .736, sd: .039. l1 in left CR and SLF: range: 7.4E4 e 9.2E4, mean: 8.2E4, sd: 3.7E5. MD in left SLF and IFOF: range: 3.9E4 e 4.7E4, mean: 4.3E4, sd: 1.8E5. One-sided Bayesian hypothesis tests for positive correlations were performed on these data. Results are shown in Kanai, Dong, et al., (2011) reported that individual differences in the degree of distractibility (CFQ) are correlated with GM volume in several brain areas. In line with the original authors' theorizing and results, we hypothesized a positive correlation between CFQ scores and GM volume in left SPL, and a negative correlation between CFQ and GM volumes in left mPFC. The following summary statistics describe our data: CFQ: range: 5e29, mean: 16.472, sd: 5.443. GM in left SPL: range: .378e.812, mean: .545, sd: .113. GM in left mPFC: range: .342e.693, mean: .499, sd: .101. Results of the one-sided Bayesian hypothesis tests for correlations are shown in Table 5 and Fig. 6. In both cases we find anecdotal support (“not worth more than a bare mention”, Jeffreys, 1961, Appendix B) for the null hypothesis. In order to provide a complete report of the SBB correlations found here in comparison with the original findings, Figs. S13e14 show posterior probability plots of these effects. The additional exploratory Bayes factor analyses with informative priors (Verhagen & Wagenmakers, 2014) show that for both effects there is anecdotal evidence in favor of the proponent's hypothesis compared to the null hypothesis. Figs. S13e14 (bottom) show posteriors for these exploratory Bayes factor analyses. All p-values indicate failed replications. Table 4 e Results of the one-sided Bayesian hypothesis tests for positive correlations. Data pair ROI BAS-Total and l1 Left CR and SLF BAS-FUN and FA Left CR and SLF BAS-FUN and l1 Left CR and SLF BAS-FUN and MD Left SLF and IFOF Confirmatory norig nrep rorig rrep BF01 Evidence cat. 51 35 .51 .28 11.74 51 36 .52 .19 9.40 51 35 .58 .24 10.57 51 36 .51 .15 2.04 Exploratory BF0r Evidence cat. p-value Strong (H0) 249.41 Extreme (H0) .948 Moderate (H0) 170.51 Extreme (H0) .861 Strong (H0) 848.06 Extreme (H0) .915 Moderate (H0) .187 Anecdotal (H0) 4.13 Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 12 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 B L1 in left CR and left SLF FA in left CR and left SLF A r = -.19 BF01 = 9.40 r= -.28 BF01= 11.74 15 20 25 30 5 6 7 Bas-Total 8 9 10 11 12 10 11 12 Bas-FUN D L1 in left CR and left SLF MD in left SLF and left IFOF C r = .15 BF01 = 2.04 r = -.24 BF01 = 10.57 5 6 7 8 9 10 11 12 5 6 Bas-FUN 7 8 9 Bas-FUN Fig. 5 e Scatterplots of replication 3: Xu et al. (2012). (A) The relationship between Bas-total and l1 in left CR and left SLF. (BeD) The relationship between Bas-FUN and (B) FA in left CR and left SLF, (C) l1 in left CR and left SLF, and (D) MD in left SLF and left IFOF. 3.5. Replication 5: Westlye et al., 2011 Westlye et al. (2011) reported that individual differences in aspects of attention (executive control and alerting) are correlated with cortical thickness in several brain areas. In line with the original authors' theorizing and results, we hypothesized negative correlations between executive control scores and CT in left caudal ACC, left STG, and right MTG. In addition, we hypothesized a negative correlation between alerting scores and CT in left SPL. One participant was excluded due to cortical thickness measures deviating more than 2.5 SDs from the group mean. After outlier rejection, the following summary statistics describe our data: Alerting: range: .068e.157, mean: .064, sd: .050. Executive control: range: .057e.402, mean: .229, sd: .082. CT in left caudal ACC: range: 2.464e2.979, mean: 2.671, sd: .121. CT in left STG: range: 2.692e3.075, mean: 2.901, sd: .083. CT in right MTG: range: 2.361e2.570, mean: 2.478, sd: .050. CT in left SPL: range: 2.116e2.610, mean: 2.360, sd: .103. One-sided Bayesian hypothesis tests for negative correlations were performed on these data. Results are shown in Table 6 and Fig. 7. In all cases we find support for the null hypothesis. The Bayes factors show that there is moderate support for the null hypothesis in one out of four tests (i.e., no correlation between Table 5 e Results of the one-sided Bayesian hypothesis tests for positive correlations. In line with the prediction of a negative correlation, the test was flipped in sign for the correlation between CFQ and GM in left mPFC. Data pair ROI Confirmatory norig CFQ and GM volume Left SPL 144 Left mPFC 144 nrep 36 36 Exploratory rorig rrep BF01 Evidence cat. BF0r Evidence cat. p-value .38 .28 .22 .19 1.24 1.51 Anecdotal (H0) Anecdotal (H0) .73 .67 Anecdotal (Hr) Anecdotal (Hr) .102 .129 Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 13 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 A B GM in left SPL GM in left lmPFC r = -.19 BF01 = 1.51 r = .22 BF01 = 1.24 5 10 15 20 25 5 10 CFQ score 15 20 25 CFQ score Fig. 6 e Scatterplots of replication 4: Kanai, Dong, et al., (2011). The relationship between CFQ score and GM in (A) left lmPFC, and (B) left SPL. alerting scores and CT in left SPL). Our data are ambiguous with regard to the correlations between executive control scores and CT in left caudal ACC, left STG, and right MTG. In order to provide a complete report of the SBB correlations found here in comparison with the original findings, Figs. S15e18 show posterior probability plots of these effects. The additional exploratory Bayes factor analyses with informative priors (Verhagen & Wagenmakers, 2014) show that for 3 effects there is anecdotal evidence in favor of the proponent's hypothesis compared to the null hypothesis. For one effect there is moderate evidence in favor of the null hypothesis. Figs. S15e18 (bottom) show posteriors for these exploratory Bayes factor analyses. All p-values indicate failed replications. 3.6. Summary of results Our results show an attenuation in effect size for almost all effects. To illustrate this overall attenuation, Fig. 8 shows the posterior probability distributions for all effects under scrutiny. Effect sizes seem to attenuate towards zero, or sometimes even shift to an opposite direction. However, for one effect from Kanai et al. (2012), the effect size is similar to the effect size found in the original study. For this effect our exploratory analyses indicate successful replications. In addition, three effects from the Westlye et al. (2011) study also show similar effect sizes to the ones found in the original investigation. For these effects, the addition of data could narrow the posterior probability distributions, potentially resulting in a successful replication. 4. Discussion In this study we set out to replicate five experiments showing SBB correlations. We adopted a preregistered, purely confirmatory approach so as to avoid common pitfalls in neuroscience such as the use of nonindependent analysis (Vul et al., 2009), double dipping (Kriegeskorte et al., 2009), obscure data collection and analysis which increase falsepositive rates (Simmons et al., 2011), and confirmation and hindsight bias on the part of the researcher (Wagenmakers et al., 2012). The five studies we attempted to replicate contained a total of 17 SBB correlations. The results from our confirmatory analyses show that we were unable to successfully replicate any of these 17 correlations. For all but one of the 17 findings under scrutiny, Bayesian hypothesis tests indicated evidence in favor of the null hypothesis. The extent Table 6 e Results of the one-sided Bayesian hypothesis tests for positive correlations. In line with the prediction of negative correlations, the tests were flipped in sign. Data pair ROI Confirmatory norig Executive control and CT left caudal ACC 132 left STG 132 right MTG 132 Alerting and CT left SPL 132 nrep Exploratory rorig rrep BF01 Evidence cat. BF0r Evidence cat. p-value 35 35 35 .21 .15 .13 .18 .14 .19 1.71 2.23 1.60 Anecdotal (H0) Anecdotal (H0) Anecdotal (H0) .67 .81 .65 Anecdotal (Hr) Anecdotal (Hr) Anecdotal (Hr) .153 .211 .141 35 .26 .16 8.58 Moderate (H0) 7.70 Moderate (H0) .824 Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 14 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 r = -.18 BF01 = 1.71 CT in left STL B CT in left caudal ACC A r = -.14 BF01 = 2.23 EC score EC score C D CT in left SPL CT in right MTL r = .16 BF01 = 8.58 r = -.19 BF01 = 1.60 .05 .15 .25 .35 EC score -.05 .00 .05 .10 .15 A score Fig. 7 e Scatterplots of replication 5: Westlye et al. (2011). (AeC) The relationship between EC scores and CT in (A) left caudal ACC, (B) left STL, and (C) right MTL. (D) The relationship between A scores and CT in left SPL. of this support ranged from anecdotal (Bayes factor < 3) to strong (Bayes factor > 10). Our additional exploratory analyses consisted of computing p-values, and a Bayes factor using an alternative method recently developed by Verhagen and Wagenmakers (2014). This method employs a more specific alternative hypothesis (termed the proponent's hypothesis), which predicts that the effect size is similar to the effect size of the original finding, rather than just predicting the direction of the effect. This analysis generally provided similar or greater support for the null hypothesis. In addition, 16 out of 17 p-values were higher than threshold (.05), indicating unsuccessful replications. For one effect in the Kanai et al. (2012), the p-value indicated a successful replication. In the current replication attempt we aimed to replicate the original experiments as closely as possible. In order to adhere to this plan we adopted a strictly confirmatory framework by publishing a ‘Methods and Analysis document’ online before any data were inspected or analyzed. This M&A document described all acquisition and analysis plans. After data analysis was complete it became clear that for some analyses, better alternative methods are available. However, the current replication attempt was strictly confirmatory, and thus we choose to (1) not perform these alternative analysis methods, and (2) make the data publicly available,1 so that other researchers might perform these alternative analysis methods instead. It should be noted, however, that these alternative analysis methods can no longer be presented as strictly confirmatory. Despite our best efforts to replicate the original experiments as closely as possible, this was partly not feasible and partly not desired. Thus, there are a number of deviations from the original study protocols. In the following section, deviations will be discussed with respect to the possibility that they contributed to spurious non-replication (i.e., a failure to detect a true correlation) of the investigated SBB correlations. 1. The sample characteristics of the present replication differed from the sample characteristics in the original studies (e.g., in terms of mean age). This might have led to systematic differences in the behavioral measures. We addressed this issue by correcting our data for age and gender, as was done in most original studies included in our replication attempt. Differences in sample 1 The data set can be freely downloaded from the NITRC Neuroimaging data repository: https://www.nitrc.org/projects/ confrep2014/. Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 15 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 Forstmann et al. (2010) individual differences in cortico-striatal tract strength predict response caution 1 2 3 Kanai et al. (2012) 4 Individual differences in grey matter density predict online and offline social network size 5 6 7 8 Xu et al. (2012) 9 Individual differences in white matter integrity predict behavioral activation scores 10 11 Kanai et al. (2011) 12 Individual differences in grey matter density predict distractibility 13 Original study 14 Westlye et al. (2011) Replication in original study 15 Individual differences in cortical thickness predict executive control and alerting aspects of attention Current replication 16 17 -1.0 -.5 0 .5 1.0 r) Fig. 8 e Summary image of our replication results. 95% confidence intervals of posterior probability distributions are shown for the original studies (red), replications within original studies (purple), and the current independent replication attempt (blue). individual effects: (1): LBA flexibility correlated to tract strength between pre-supplementary motor area and striatum. (2e6): FBN correlated to grey matter volume in (2) left middle temporal gyrus, (3) right superior temporal sulcus, (4) right entorhinal cortex, (5) left amygdala, and (6) right amygdala. (7) SNS correlated to grey matter volume in right amygdala. (8) BAS-total correlated to l1 in left CR and SLF. (9) BAS-FUN correlated to FA in left CR and SLF. (10) BAS-FUN correlated to l1 in left CR and SLF. (11) BAS-FUN correlated to MD in left SLF and IFOF. (12e13) CFQ correlated to grey matter volume in (12) left superior parietal lobe and (13) left middle prefrontal cortex. (14e16) Executive control correlated to cortical thickness in (14) left caudal anterior cingulate cortex, (15) left superior temporal gyrus, and (16) right middle temporal gyrus. (17) Alerting correlated to cortical thickness in left superior parietal lobe. characteristics might still have non-linear effects on our measures, or aging might have differing effects on different brain regions. Future replication studies could take into account the characteristics of the sample used in the original study, and attempt to match participants in the replication sample to participants in the original sample more closely. Despite the relevance of this concern, note that in cognitive neuroscience, one often makes claims with regard to a population of humans (i.e., generalizing towards an ‘average person’). If the reported effects are indeed non-specific to the sample and its characteristics, there is no reason to assume a priori that a sample with different characteristics impairs our ability to detect the effect. For this reason we chose to acquire Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 16 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 data from the current sample, and hypothesize effects as they were described in the original studies. In order to address the concern that (non-linear) effects of differences in sample characteristics might still impair our ability to find these effects, additional research is needed to investigate the specific sample characteristics for which these effects are present. Similarly, our data differ from the data in the original studies, for instance in terms of the spread of some of the behavioral measures. However, these differences should have little impact on the correlational analyses, since these are not based on the values of the two measures of interest, but on their linear dependence. Only one behavioral measure (i.e., scores on the political orientation questionnaire) did not show enough variance in order to perform a replication attempt. With respect to sample size, it should be noted that while our sample size was lower than most original studies, our results showed that in our data set, 8 out of a total of 17 hypothesized effects were contradicted with moderate or strong levels of evidence. Thus, even though larger samples are always better than smaller samples from a pre-experimental perspective, our Bayesian post-experimental perspective shows that even with 36 participants it is possible to obtain informative results.2 Nevertheless, we encourage additional replication attempts of SBB correlations using larger sample sizes in order to further decrease uncertainty about the replicability of these effects. 2. The MRI data used in the present replication were acquired using a different scanner and with slightly different scanning parameter settings than the MRI data of the original studies. However, recent multi-site reliability studies have shown that these differences have only little impact in both VBM/CT (Jovicich et al., 2013; Schnack et al., 2010) and DTI analyses (Fox et al., 2012). 3. In our TBSS analysis pipeline, another addition to the original protocols is the registration of the ROI spatial maps to our mean FA skeleton. We used spatial maps that were provided by the original authors, and comprised those voxels that correlated with the behavioral measure in the original study. As opposed to using comparably large atlas-based ROI, this approach minimizes the probability that the contribution of a small subset of voxels that correlate with the behavioral measure is canceled out due to averaging across all voxels within the atlas-based ROI. However, in order to be able to use the spatial maps from the original studies we had to register them into the skeleton space common to all participants in our sample. Following the principle of parsimony, we used affine-only (linear) registration with 12 degrees-of-freedom (DoF), which does not guarantee perfect alignment of even the major tracts (Smith et al., 2006). Residual misalignments would be reduced with the use of nonlinear registration. 2 In general, it is possible for low-power experiments to yield diagnostic results, and for high-power experiments to yield nondiagnostic results. By conditioning on the observed data, Bayes factors quantify the evidential impact of the information at hand, ignoring hypothetical outcomes that did not occur (Wagenmakers et al., in press; http://ejwagenmakers.com/ inpress/APowerFallacy.pdf). However, such high-DoF alternatives might warp the images so much that the overall structure is not preserved (Smith et al., 2006). It should be noted that, due to the residual misalignments from the linear registration, only a subset of the voxels contained in the registered spatial maps was used in the correlational analysis. Only voxels, overlapping with the mean FA skeleton were considered. The reduction in the size of ROI would be a concern if we had performed voxelwise statistics (Smith et al., 2006). However, since we aggregated only one value per ROI, it is unlikely that the smaller ROIs have led to spurious nonreplication. On a more general note, software packages may differ slightly in the statistical methods that they employ. These differences can have a relevant impact on the results (e.g., Gronenschild et al., 2012; Rajagopalan, Yue, & Pioro, 2014). Our data are publicly available, so that other researchers can carry out additional analyses to probe the robustness of our results. However, such analyses can only be partly confirmatory. Here we restrict ourselves to reporting preregistered, purely confirmatory analyses performed in FSL (Douaud et al., 2007). 4. The use of Bayesian hypothesis tests for correlations instead of the common null hypothesis significance tests was motivated by two compelling advantages. First, unlike p values the Bayes factor can quantify evidence in favor of the null hypothesis. Second, unlike p-values, Bayes factors do not have the tendency to over-estimate the evidence against the null hypothesis (Edwards et al., 1963; Sellke, Bayarri, & Berger, 2001; Wetzels et al., 2011). Note, however, that we have included p-values as exploratory tests. Another deviation concerning the correlational analyses is that we used one-sided instead of two-sided tests, incorporating our prior expectations about the direction of the SBB correlations based on the findings of the original studies. However, this approach provides more compelling evidence (e.g., Hoijtink, Klugkist, & Boelen, 2008; Wagenmakers, Lodewyckx, Kuriyal, & Grasman, 2010) and should facilitate replication of true SBB correlations and not contribute to spurious non-replication. 5. While our ROI approach is specific with regard to the location at which we predict the SBB correlation, it does not take anatomical variability between data sets into account. In addition, we extracted the mean signal from the ROIs instead of performing voxel-wise correlations within the ROIs. This process, in combination with anatomical variability between data sets, introduces noise into the structural measures, potentially concealing the SBB correlation. Future replication work might employ different approaches, which take into account potential anatomical variability, while still making clear predictions with regard to spatial locations of SBB correlations. Note that this point emphasizes the importance of replications within the current field of work. Given that there is random variation in the location of the effect as well as the size of the effect, replication studies are necessary in order to identify the precise location of the effect in addition to the precise effect size. Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 17 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 From the above discussion, one might be tempted to conclude that most of the SBB correlations tested here simply may not exist. However, as previously mentioned, a single replication cannot be conclusive in terms of confirmation or refutation of a finding. We acknowledge the recent replication efforts within the social sciences in general and psychology and neuroscience in particular; an excellent example is the Reproducibility Project of the Open Science Framework (http://openscienceframework.org/) and the first Registered Replication Report (Alogna et al., 2014). Still, to our knowledge, the present replication is the first independent attempt to replicate SBB correlations, despite the considerable number of publications on the matter. We believe that in order to establish correlations between behavior and structural properties of the brain more firmly, it is desirable for the field to replicate SBB correlations, preferably using preregistration protocols and Bayesian inference methods. Acknowledgments We would like to acknowledge the authors of the articles we attempted to replicate for sharing their spatial maps, without which we would not have been able to conduct this research. Supplementary data Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.cortex.2014.11.019. Appendix A. Replication Bayes Factor A replication Bayes factor (Verhagen & Wagenmakers, 2014) answers the question: “Is the effect from the replication attempt comparable to what was found before, or is it absent?”. When a correlational study is replicated, the replication Bayes factor compares evidence in favor of the null hypothesis of no effect, H0: r ¼ 0, with the evidence in favor of the alternative hypothesis that the effect is equal to the effect found in the original study, H1: r ~ “posterior distribution from the original study” The replication Bayes factor is calculated in two steps: In the first step the posterior distribution of the original study is obtained, assuming a uniform prior on the correlation. The density of this posterior distribution was given by Jeffreys (1961, pp 175, Equation 9), and simplifies: ð1 1 ðn 1Þ r2 Þ2 ð1 rrÞ    p rYorig ¼ Z ð1 ð1 n 3 2 1 ðn r2 Þ2 n rrÞ 1Þ 3 2 ppffiffi Gðn 1Þ 2 G n rffiffiffi p Gðn 2 G n 2 F1 1 1 ; 2 2 þ 12 rr 1 2 1Þ 1 2 2 F1 1 1 1 ; þ rr dr 2 2 2 where 2F1 is Gauss' hypergeometric function (Abramowitz & Stegun, 1970, sec. 15). The second step consists of the computation of the Bayes factor by integration over this posterior distribution: B10 ¼ Z ð1 Z ð1 ð1 ð1 PðYjH1 Þ PðYjH0 Þ 1 ðn r2 Þ2 n 1Þ 3 2    p rYorig dd rrÞ    p r ¼ 0Yorig 1 ðn r2 Þ2 n rrÞ 1Þ 3 2    p rYorig dd which can be done by performing a one-dimensional integration. R code to perform this analysis can be found in this link: http://www.josineverhagen.com/?page_id¼76. references Abramowitz, M., & Stegun, I. (1970). Handbook of mathematical functions. New York: Dover Publications.  Birch, S., Alogna, V. K., Attaya, M. K., Aucoin, P., Bahnı́k, S., Birt, A., et al. (2014). Registered Report: Schooler and EngstlerSchooler (1990). Perspectives on Psychological Science, 9(5), 556e578. Banissy, M. J., Kanai, R., Walsh, V., & Rees, G. (2012). Interindividual differences in empathy are reflected in human brain structure. NeuroImage, 62(3), 2034e2039. Behrens, T. E. J., Johansen-Berg, H., Woolrich, M. W., Smith, S. M., Wheeler-Kingshott, C. A. M., Boulby, P. A., et al. (2003). Noninvasive mapping of connections between human thalamus and cortex using diffusion imaging. Nature Neuroscience, 6(7), 750e757. Bickart, K. C., Wright, C. I., Dautoff, R. J., Dickerson, B. C., & Barrett, L. F. (2011). Amygdala volume and social network size in humans. Nature Neuroscience, 14(2), 163e164. Britten, K. H., Shadlen, M. N., Newsome, W. T., & Movshon, J. A. (1992). The analysis of visual motion: a comparison of neuronal and psychophysical performance. The Journal of Neuroscience, 12(12), 4745e4765. Broadbent, D. E., Cooper, P. F., FitzGerald, P., & Parkes, K. R. (1982). The cognitive failures questionnaire (CFQ) and its correlates. British Journal of Clinical Psychology, 21(1), 1e16. Brown, S. D., & Heathcote, A. (2008). The simplest complete model of choice response time: linear ballistic accumulation. Cognitive Psychology, 57(3), 153e178. Button, K. S., Ioannidis, J. P. A., Mokrysz, C., Nosek, B. A., Flint, J., Robinson, E. S. J., et al. (2013). Power failure: why small sample size undermines the reliability of neuroscience. Nature Reviews Neuroscience, 14(5), 365e376. Campbell-Meiklejohn, D. K., Kanai, R., Bahrami, B., Bach, D. R., Dolan, R. J., Roepstorff, A., et al. (2012). Structure of orbitofrontal cortex predicts social influence. Current Biology : CB, 22(4), R123eR124. Carver, C. S., & White, T. L. (1994). Behavioral inhibition, behavioral activation, and affective responses to impending reward and punishment: the BIS/BAS Scales. Journal of Personality and Social Psychology, 67(2), 319e333. Chambers, C. D. (2013). Registered reports: a new publishing initiative at Cortex. Cortex, 49(3), 609e610. Cohen, S. (1997). Social ties and susceptibility to the common cold. JM&A: The Journal of the American Medical Association, 277(24), 1940. Dale, A. M., Fischl, B., & Sereno, M. I. (1999). Cortical surface-based analysis. I. Segmentation and surface reconstruction. NeuroImage, 9(2), 179e194. Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 18 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 Dale, A. M., & Sereno, M. I. (1993). Improved Localization of cortical Activity by combining EEG and MEG with MRI cortical surface reconstruction: a linear approach. Journal of Cognitive Neuroscience, 5(2), 162e176. Davis, M. H. (1980). A multidimensional approach to individual differences in empathy. JSAS Catalog of Selected Documents in Psychology, 10, 85e103. Dienes, Z. (2008). Understanding psychology as a Science: An introduction to scientific and statistical inference. New York: Palgrave MacMillan. Douaud, G., Smith, S., Jenkinson, M., Behrens, T., JohansenBerg, H., Vickers, J., et al. (2007). Anatomically related grey and white matter abnormalities in adolescent-onset schizophrenia. Brain : A Journal of Neurology, 130(Pt 9), 2375e2386. Edwards, W., Lindman, H., & Savage, L. J. (1963). Bayesian statistical inference for psychological research. Psychological Review, 70(3), 193e242. Fan, J., McCandliss, B. D., Sommer, T., Raz, A., & Posner, M. I. (2002). Testing the efficiency and independence of attentional networks. Journal of Cognitive Neuroscience, 14(3), 340e347. Fischl, B., & Dale, A. M. (2000). Measuring the thickness of the human cerebral cortex from magnetic resonance images. Proceedings of the National Academy of Sciences of the United States of America, 97(20), 11050e11055. Fischl, B., van der Kouwe, A., Destrieux, C., Halgren, E.,
gonne, F., Salat, D. H., et al. (2004). Automatically Se parcellating the human cerebral cortex. Cerebral Cortex, 14(1), 11e22. Fischl, B., Liu, A., & Dale, A. M. (2001). Automated manifold surgery: constructing geometrically accurate and topologically correct models of the human cerebral cortex. IEEE Transactions on Medical Imaging, 20(1), 70e80. Fischl, B., Salat, D. H., Busa, E., Albert, M., Dieterich, M., Haselgrove, C., et al. (2002). Whole brain segmentation: automated labeling of neuroanatomical structures in the human brain. Neuron, 33(3), 341e355. Fischl, B., Salat, D. H., van der Kouwe, A. J. W., Makris, N.,
gonne, F., Quinn, B. T., et al. (2004). Sequence-independent Se segmentation of magnetic resonance images. NeuroImage, 23(Suppl 1), S69eS84. Fischl, B., Sereno, M. I., & Dale, A. M. (1999). Cortical surface-based analysis. II: Inflation, flattening, and a surface-based coordinate system. NeuroImage, 9(2), 195e207. Fischl, B., Sereno, M. I., Tootell, R. B., & Dale, A. M. (1999). Highresolution intersubject averaging and a coordinate system for the cortical surface. Human Brain Mapping, 8(4), 272e284. Fleming, S. M., Weil, R. S., Nagy, Z., Dolan, R. J., & Rees, G. (2010). Relating introspective accuracy to individual differences in brain structure. Science, 329(5998), 1541e1543. € fer, A., Neumann, J., Forstmann, B. U., Anwander, A., Scha Brown, S., Wagenmakers, E.-J., et al. (2010). Cortico-striatal connections predict control over speed and accuracy in perceptual decision making. Proceedings of the National Academy of Sciences, 107(36), 15916e15920. Fox, R. J., Sakaie, K., Lee, J. C., Debbins, J. P., Liu, Y., Arnold, D. L., et al. (2012). A validation study of multicenter diffusion tensor imaging: reliability of fractional anisotropy and diffusivity values. American Journal of Neuroradiology, 33(4), 695e700. Goldacre, B. (2009). Bad science. London: Fourth Estate. Gold, J. I., & Shadlen, M. N. (2007). The neural basis of decision making. Annual Review of Neuroscience, 30, 535e574. Good, C. D., Johnsrude, I. S., Ashburner, J., Henson, R. N., Friston, K. J., & Frackowiak, R. S. (2001). A voxel-based morphometric study of ageing in 465 normal adult human brains. NeuroImage, 14(1 Pt 1), 21e36. Graham, J., Haidt, J., & Nosek, B. A. (2009). Liberals and conservatives rely on different sets of moral foundations. Journal of Personality and Social Psychology, 96(5), 1029e1046. Gronenschild, E. H., Habets, P., Jacobs, H. I., Mengelers, R., Rozendaal, N., van Os, J., et al. (2012). The effect of FreeSurfer version, workstation type, and Macintosh operating system version on anatomical volume and cortical thickness measurements. PLoS One, 7(6). Groot, A. D. (1969). Methodology: Foundations of inference and research in the behavioral sciences. The Hague: Mouton. Han, X., Jovicich, J., Salat, D., van der Kouwe, A., Quinn, B., Czanner, S., et al. (2006). Reliability of MRI-derived measurements of human cerebral cortical thickness: the effects of field strength, scanner upgrade and manufacturer. NeuroImage, 32(1), 180e194. Hoijtink, H., Klugkist, I., & Boelen, P. A. (2008). An introduction to Bayesian evaluation of informative hypotheses (pp. 1e3). New York, NY: Springer (New York). Ioannidis, J. P. A. (2005). Why most published research findings are false. PloS Med, 2(8). Ioannidis, J. P. A. (2012). Why science is not necessarily selfcorrecting. Perspectives on Psychological Science, 7(6), 645e654. Jeffreys, H. (1961). Theory of probability. Oxford, UK: Oxford University Press. Jenkinson, M., & Smith, S. (2001). A global optimisation method for robust affine registration of brain images. Medical Image Analysis, 5(2), 143e156. John, L. K., Loewenstein, G., & Prelec, D. (2012). Measuring the prevalence of questionable research practices with incentives for truth telling. Psychological Science, 23(5), 524e532. Jovicich, J., Czanner, S., Greve, D., Haley, E., van der Kouwe, A., Gollub, R., et al. (2006). Reliability in multi-site structural MRI studies: effects of gradient non-linearity correction on phantom and human data. NeuroImage, 30(2), 436e443.
sJovicich, J., Marizzoni, M., Sala-Llonch, R., Bosch, B., Bartre Faz, D., Arnold, J., et al. (2013). Brain morphometry reproducibility in multi-center 3T MRI studies: a comparison of cross-sectional and longitudinal segmentations. NeuroImage, 83, 472e484. Kanai, R., Bahrami, B., & Rees, G. (2010). Human parietal cortex structure predicts individual differences in perceptual rivalry. Current Biology, 20(18), 1626e1630. Kanai, R., Bahrami, B., Roylance, R., & Rees, G. (2012). Online social network size is reflected in human brain structure. Proceedings. Biological Sciences/the Royal Society, 279(1732), 1327e1334. Kanai, R., Carmel, D., Bahrami, B., & Rees, G. (2011). Structural and functional fractionation of right superior parietal cortex in bistable perception. Current Biology, 21(3), 106e107. Kanai, R., Dong, M. Y., Bahrami, B., & Rees, G. (2011). Distractibility in daily life is reflected in the structure and function of human parietal cortex. Journal of Neuroscience, 31(18), 6620e6626. Kanai, R., Feilden, T., Firth, C., & Rees, G. (2011). Political orientations are correlated with brain structure in young adults. Current Biology: CB, 21(8), 677e680. Kanai, R., & Rees, G. (2011). The structural basis of inter-individual differences in human behavior and cognition. Nature Reviews Neuroscience, 12(4), 231e242. Kass, R. E., & Raftery, A. E. (1995). Bayes factors. Journal of the American Statistical Association, 90, 773e795. King, A. V., Linke, J., Gass, A., Hennerici, M. G., Tost, H., Poupon, C., et al. (2012). Microstructure of a three-way anatomical network predicts individual differences in response inhibition: a tractography study. NeuroImage, 59(2), 1949e1959. Kriegeskorte, N., Simmons, W. K., Bellgowan, P. S. F., & Baker, C. I. (2009). Circular analysis in systems neuroscience: the dangers of double dipping. Nature Neuroscience, 12(5), 535e540. Lee, M. D., & Wagenmakers, E. J. (2013). Bayesian modeling for cognitive science: A practical course. Cambridge: Cambridge University Press. Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019 c o r t e x x x x ( 2 0 1 5 ) 1 e1 9 Lewis, G. J., Kanai, R., Bates, T. C., & Rees, G. (2012). Moral values are associated with individual differences in regional brain volume. Journal of Cognitive Neuroscience, 24(8), 1657e1663. MacArthur, D. (2012). Methods: face up to false positives. Nature, 487(7408), 427e428. Pashler, H., & Harris, C. R. (2012). Is the replicability crisis overblown? Three arguments examined. Perspectives on Psychological Science, 7(6), 531e536. Pashler, H., & Wagenmakers, E.-J. (2012). Editors' introduction to the special section on replicability in psychological science a crisis of confidence? Perspectives on Psychological Science, 7(6), 528e530. Rajagopalan, V., Yue, G. H., & Pioro, E. P. (2014). Do preprocessing algorithms and statistical models influence voxel-based morphometry (VBM) results in amyotrophic lateral sclerosis patients? A systematic comparison of popular VBM analytical methods. Journal of Magnetic Resonance Imaging, 40(3), 662e667. Reuter, M., Rosas, H. D., & Fischl, B. (2010). Highly accurate inverse consistent registration: a robust approach. NeuroImage, 53(3), 1181e1196. Reuter, M., Schmansky, N. J., Rosas, H. D., & Fischl, B. (2012). Within-subject template estimation for unbiased longitudinal image analysis. NeuroImage, 61(4), 1402e1418. Rosenthal, R. (1979). The file drawer problem and tolerance for null results. Psychological Bulletin, 86(3), 638e641. Rouder, J. N., Morey, R. D., Speckman, P. L., & Province, J. M. (2012). Default Bayes factors for ANOVA designs. Journal of Mathematical Psychology, 56(5), 356e374. Rouder, J. N., Speckman, P. L., Sun, D., Morey, R. D., & Iverson, G. (2009). Bayesian t tests for accepting and rejecting the null hypothesis. Psychonomic Bulletin & Review, 16(2), 225e237. Schnack, H. G., van Haren, N. E. M., Brouwer, R. M., van Baal, G. C. M., Picchioni, M., Weisbrod, M., et al. (2010). Mapping reliability in multicenter MRI: Voxel-based morphometry and cortical thickness. Human Brain Mapping, 31(12), 1967e1982.
gonne, F., Dale, A. M., Busa, E., Glessner, M., Salat, D., Se Hahn, H. K., et al. (2004). A hybrid approach to the skull stripping problem in MRI. NeuroImage, 22(3), 1060e1075.
gonne, F., Pacheco, J., & Fischl, B. (2007). Geometrically accurate Se topology-correction of cortical surfaces using nonseparating loops. IEEE Transactions on Medical Imaging, 26(4), 518e529. Sellke, T., Bayarri, M. J., & Berger, J. O. (2001). Calibration of r values for testing precise null hypotheses. The American Statistician, 55(1), 62e71. Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2011). Falsepositive psychology undisclosed flexibility in data collection and analysis allows presenting anything as significant. Psychological Science, 22(11), 1359e1366. Sled, J. G., Zijdenbos, A. P., & Evans, A. C. (1998). A nonparametric method for automatic correction of intensity nonuniformity in MRI data. IEEE Transactions on Medical Imaging, 17(1), 87e97. Smith, S. M. (2002). Fast robust automated brain extraction. Human Brain Mapping, 17(3), 143e155. Smith, S. M., Jenkinson, M., Johansen-Berg, H., Rueckert, D., Nichols, T. E., Mackay, C. E., et al. (2006). Tract-based spatial statistics: voxelwise analysis of multi-subject diffusion data. NeuroImage, 31(4), 1487e1505. Smith, S. M., Jenkinson, M., Woolrich, M. W., & Beckmann, C. F. (2004). Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage, 23(Suppl 1), S208eS219. 19 Stileman, E., & Bates, T. (2007). Construction of the social network score (SNS) questionnaire for undergraduate students, and an examination of the pre-requisites for large social networks in humans. Unpublished undergraduate thesis. See http://hdl. handle.net/1842/2553. Tuch, D. S., Salat, D. H., Wisco, J. J., Zaleta, A. K., Hevelone, N. D., & Rosas, H. D. (2005). Choice reaction time performance correlates with diffusion anisotropy in white matter pathways supporting visuospatial attention. Proceedings of the National Academy of Sciences of the United States of America, 102(34), 12212e12217. Verhagen, A. J., & Wagenmakers, E.-J. (2014). Bayesian tests to quantify the result of a replication attempt. Journal of Experimental Psychology: General, 143, 1457e1475. Vul, E., Harris, C., Winkielman, P., & Pashler, H. (2009). Puzzlingly high correlations in fMRI studies of emotion, personality, and social cognition. Perspectives on Psychological Science, 4(3), 274e290. Wagenmakers, E. J. (2007). A practical solution to the pervasive problems of p values. Psychonomic Bulletin & Review, 14(5), 779e804. Wagenmakers, E.-J., Lodewyckx, T., Kuriyal, H., & Grasman, R. (2010). Bayesian hypothesis testing for psychologists: a tutorial on the Savage-Dickey method. Cognitive Psychology, 60(3), 158e189. Wagenmakers, E.-J., Verhagen, J., Ly, A., Bakker, M., Lee, M. D., Matzke, D., et al. (2015). A power fallacy. Behavior Research Methods (in press). Wagenmakers, E.-J., Wetzels, R., Borsboom, D., & van der Maas, H. L. J. (2011). Why psychologists must change the way they analyze their data: the case of psi: comment on Bem (2011). Journal of Personality and Social Psychology, 100(3), 426e432. Wagenmakers, E.-J., Wetzels, R., Borsboom, D., van der Maas, H. L. J., & Kievit, R. A. (2012). An agenda for purely confirmatory research. Perspectives on Psychological Science, 7(6), 632e638. Wallace, J. C., Kass, S. J., & Stanny, C. J. (2002). The cognitive failures questionnaire revisited: dimensions and correlates. The Journal of General Psychology, 129(3), 238e256. Wallach, H., & O'Connel, D. N. (1953). The kinetic depth effect. Journal of Experimental Psychology, 45(4), 205e217. Westlye, L. T., Grydeland, H., Walhovd, K. B., & Fjell, A. M. (2011). Associations between regional cortical thickness and attentional networks as measured by the attention network test. Cerebral Cortex, 21(2), 345e356. Wetzels, R., Matzke, D., Lee, M. D., Rouder, J. N., Iverson, G. J., & Wagenmakers, E. J. (2011). Statistical evidence in experimental psychology: an empirical comparison using 855 t tests. Perspectives on Psychological Science, 6(3), 291e298. Wetzels, R., & Wagenmakers, E.-J. (2012). A default Bayesian hypothesis test for correlations and partial correlations. Psychonomic Bulletin & Review, 19(6), 1057e1064. Wolfe, J. M. (2013). Registered reports and replications in attention, perception, & psychophysics. Attention, Perception, & Psychophysics, 75(5), 781e783. Xu, J., Kober, H., Carroll, K. M., Rounsaville, B. J., Pearlson, G. D., & Potenza, M. N. (2012). White matter integrity and behavioral activation in healthy subjects. Human Brain Mapping, 33(4), 994e1002. Please cite this article in press as: Boekel, W., et al., A purely confirmatory replication study of structural brain-behavior correlations, Cortex (2015), http://dx.doi.org/10.1016/j.cortex.2014.11.019