COMPARING CHANGE IN NEURAL INDICES OF ATTENTIONAL BIAS AND COGNITIVE CONTROL BETWEEN COGNITIVE BEHAVIORAL THERAPY AND ATTENTION BIAS MODIFICATION FOR SOCIAL ANXIETY DISORDER By Courtney C. Louis A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Psychology Master of Arts 2019 ABSTRACT COMPARING CHANGE IN NEURAL INDICES OF ATTENTIONAL BIAS AND COGNITIVE CONTROL BETWEEN COGNITIVE BEHAVIORAL THERAPY AND ATTENTION BIAS MODIFICATION FOR SOCIAL ANXIETY DISORDER By Courtney C. Louis Cognitive models of S ocial Anxiety Disorder (SAD) state that an imbalance between attention bias (AB) and cognitive control (CC) maintain anxious symptoms . Event - related potentials (ERPs) serve as a precise index to assess these processes. The current study aimed to examine change in the P1 and N2 as indices of AB and CC within Cognitive Behavioral Therapy (CBT) and Attention Bias Modification (ABM) in individ uals diagnosed with SAD . The re were t w o primary aims (1) to examine change in the P1 and N2 to the faces and probe in the dot probe task , and (2) to examine whether changes in ERPs predict symptom change. The sample consisted of 50 adult patients diagnosed with SAD , who were randomly assigned to CBT ( n = 33 ) or ABM ( n = 17 ). For the first aim, multi - level models (MLM) were used to estimate growth curves for each ERP. The results revealed an increase in the N2 to the faces over time for both groups ( b = - .38, p = .02) at post - treatment. No other ERP changes reached significance (all p .09 ) . For the second aim, individual growth slopes of ERPs and symptoms were correlated with each other , however there were no relationships between ERP s and symptom change (all p .4 0 ) . Lagged MLMs with ERPs as predictors of symptoms demonstrated that a smaller N2 a t one assessment related to mor e avoidance symptoms at the next assessment for the ABM group only ( b = - 1.23, p = .02) . These results signify the need for reliable met hods to assess AB , as well as increased specificity of CC processes that are targeted and exerc ised within treatment contexts iii ACKNOWLEDGEMENTS The author would like to thank Dr. Jason Moser for his dedicated time , advisement, and support , Dr. Jonathan Huppert fo r the use this data as well as his guidance on the analyses , Dr. Katy Thakkar for her time and feedback, Dr. Yogev Kivit y for his help with the analysis strategy, the participants for their participation, and friends and family for providing support and laughter. iv TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ........................... v LIST OF FIGURES ................................ ................................ ................................ ....................... vi INTRODUCTION ................................ ................................ ................................ ........................... 1 Attentional bias and cognitive control in SAD ................................ ................................ .... 5 Event - related potentials (ERPs) as markers of attention bias and cognitive control ........... 6 Treatments for SAD: effects on attention bias and cognitive control ................................ . 7 The Present Study ................................ ................................ ................................ .............. 10 METHOD ................................ ................................ ................................ ................................ ...... 13 Participants ................................ ................................ ................................ ......................... 13 Measures ................................ ................................ ................................ ............................ 13 Treatments ................................ ................................ ................................ ......................... 18 Procedure ................................ ................................ ................................ .......................... 19 Data Analytic Approach ................................ ................................ ................................ ... 19 RESULTS ................................ ................................ ................................ ................................ ...... 23 Face - locked Growth Models ................................ ................................ .............................. 23 Probe - locked Growth Models Treatments ................................ ................................ ......... 24 Relationship between ERP change and symptom change ................................ ................. 25 Multilevel Models with Lagged Pre - Treatment Centered ERPs to Face Presentation as Predictors ................................ ................................ ................................ ........................... 26 Multilevel Models with Lagged Pre - Treatment Centered ERPs to Probe Presentation as Predictors ................................ ................................ ................................ ........................... 28 DISCUSSION ................................ ................................ ................................ ................................ 30 APPENDICES ................................ ................................ ................................ ............................... 36 APPENDIX A: Tables ................................ ................................ ................................ ....... 37 APPENDIX B: Figures ................................ ................................ ................................ ...... 44 REFERENCES ................................ ................................ ................................ .............................. 51 v LIST OF TABLES Table 1 : Descriptive Statistics on the full sample .. .. . 38 Table 2 : Final model for N2 change to face presentation in the Dot Probe Task (Pre - Treatment 8 week mark) 38 Table 3 : Final model for N2 change to face presentation in the Dot Probe Task (Pre - Treatment Post Treatment) Table 4 : Simple Model for P1 Change Over Time to Face Presentation for Slope Extraction..... 39 Table 5 : Simple Model for N2 Change Over Time to Face Presentation for Slope Extraction. ...40 Table 6 : Simple Model for N1 Change Over Time to Probe Presentation for Slope Extraction ...40 Table 7 : Simple Model for N2 Change Over Time to Probe Presentation for Slope Extraction .. . 41 Table 8 : Simple Model for LSAS Total Scores Change Time for the Slope Extraction Table 9 : Simple Model for LSAS Anxiety Scores Change Time for the Slope Extraction Table 10 : Simple Model for LSAS Avoidance Scores Change Time for the Slop e Extraction Table 11 : Table 12 : Bivariate Correlations between ERP and Symptom Slopes 43 vi LIST OF FIGURES Figure 1 : Flow Chart of Study Design . 45 Figure 2: Topographic representation of the P1 time - locked to the presentation of the faces including all observations for the final sample in the 80 - . . 46 Figure 3: The effect of Site across time for the N2 to face presentation in the dot probe task. The N2 was largest at site Fz . 47 Figure 4: The N2 to the faces over time collapsed across both groups. Increased N2 was observed across treatment. .. 47 Figure 5: Line Graph of N2 to the faces. The topographic representation depicts the N2 change difference (Pre - Treatment - Post Treatment). . . 48 Figure 6: The relationship between the slopes (pre - post treatment) for the P1 and the N2 to face presentation. . .. 48 Figure 7: The relationship between the slopes (pre - post treatment) for the P1 to face presentation and the N1 to probe presentation. .. 49 Figure 8: The relationship the N2 to face presentation and LSAS Avoidance scores, showing that a smal ler N2 predicted less symptoms for the ABM group while there was no relationship in the CBT group 49 Figure 9: The relationship the N2 to face presentation and SPIN scores, showing that a smaller N2 predicted higher symptom s for the CBT group while there was no relationship in the ABM group 50 1 INTRODUCTION Social A nxiety D isorder (SAD), characterized by an intense fear of being negatively evaluated in social situations , affect s approximately 12% of the general population (Association, 2013; Kessler, Petukhova, Sampson, Zaslavsky, & Wittchen, 2012) . Cognitive models of anxiety have implicated attentional bias toward threat and impaired cognitive control as k ey factors that maintain symptoms of SAD ( Clark, 2005; Bar - Haim et al., 2007; Eysenck, Derakshan, Santos, & Calvo, 2007; Mogg & Bradley, 201 6 ) . As such, t reatment interventions have aimed to target these factors to reduce distress and impairment for individuals with SAD . H owever , very little research has probed the effects of these treatments on neur al mechanisms of attentional bias towards threat and impaired cognitive control . E ven less has directly compared such effects across two active treatments. The current study aim s to address these gaps in the literature by examining changes in neurophysiologic al markers of attention toward threat and cognitive control following Cognitive Behavioral Therapy (CBT) and Attention Bias Modification (ABM) for SAD . Secondarily, the current study aims to assess if neurophysiological change within each treatment relates to symptom change . Attention al b ias and c ognitive c ontrol in SAD An imbalance between the salience - driven (i.e., attention bias) and goal - directed attention (i.e., cognitive control) systems has been proposed to contribute to the maintenance of anxiety disorders (Eysenck et al., 2 007 ; Mogg & Braddley, 2016 ) . Specifically , anxiety is characterized by early hypervigilance toward threatening stimuli that impairs cognitive control to efficiently complete task s . As noted by Eysenck et al. (2007) , the goal - directed attention system is defined as a variety of functions that support goal - directed behavior. These functions include the ability to inhibit dis tractors, to integrate information across a range of modalities (both internal and 2 external) , and to recruit more control in the presence of conflict ( Eysenck et al., 2007; Miller & Cohen, 2001 ; Mogg & Braddley, 2017 ) . A nxious individuals tend to experience biased attention to salient stimuli that reduces the ability to use cognitive control. As such, analyzing the disruption of these two cognitive systems and their treatment is vital for an enhanced understanding of the functional mechanisms that maintain anxiety, and their ability to be targeted throughout the course of treatment. One of the most widely utilized tasks to ass ess attentional bias within anxiety is the dot probe paradigm ( MacLeod, Mathews, & Tata, 1986) . In this task, a fixation cross is presented at the center of the screen, followed by the simultaneous presentation of two faces (i.e., threatening and non - threatening) for 500 milliseconds (ms) . Next, a probe arrow ) replaces one of the images, and participants are required to quickly and accurately identify the probe. Attentio n bias scores are derived by c omparing reaction times (RTs) on congruent trials (when the probe replaces the threatening face) with RTs on incongruent trials (when the probe replaces the neutral face). A difference score is calculated by subtracting RT s on congruent trials from incongruent trials , with positive values interpreted as an attentional bias toward threat . It is assumed that individuals have faster RTs on congruent trials because their attention was already allocated toward the threatening image . Past research has extensively studied attentional biases within anxiety, providing evidence for increased attentional bias to threat (Bar - Haim, Lamy, Pergamin, Bakermans - Kranenburg, & van Ijzendoor, 2007) . Individuals with SAD specifically experience a heightened attentional bias to stimuli that could lead to negative evaluation, such as disgust and contempt faces ( Moser , Huppert , Duval, & Simmons, 2008; Mogg, Philippot, & Bradley, 2004; Pyshar et al., 2004 ) and negative evaluative words (Amir et al., 199 6 ; Amir, Elias, Klumpp, & Przeworski, 3 2003) . Examining the processing of threatening faces can be particularly informative because they represent precisely the set of stimuli that make individuals with SAD anxious. Not surprisingly, i n a review of 74 studies examining face processing in social anxiety , Staugaard (2010) found that individuals with SAD show ed an early attentional bias toward threatening stimuli, although results become inconsistent for longer stimulus presentations (> 500 ms). For the dot probe task specifically, Bantin, Stevens, Grlach, & Hermann (2016) review ed 10 studies and found that those with SAD evidenced early attentional bias es toward threatening faces . Therefore, it is generally accepted that individuals with SAD show early attentional bias toward threatening faces . Research also suggests that anxiety is characterized by an inability to disengage from threatening cues (i.e., delayed disengagement) (Fox, Russo, Bowles, & Dutton, 2001; Yie nd & Mathews, 2001) . That is, once threatening stimul i are attended to, it is difficult for anxious individuals to disengage , and execute goal - directed behavior. This is, indeed, the case for socially anxious individuals speci fic ally . For instance, Amir, Elias, Klumpp, & Przeworski (2003) found that individuals with clinical levels of social anxiety experience delayed disengagement from social ly - evaluative words on a spatial cueing task. T h ose with SAD were slower at responding to probes when presented at a different spatial location than threatening words . Additionally , Moriya and Tanno (2011) found that those with high levels of social anxi ety (in comparison to those with low levels) were slower at responding to go trials in a modified version of the Go/No - Go task following centrally presented threatening image s . The same effect was not observed for neutral faces , implying that socially anxious individuals experienced difficulty disengaging from t hreatening faces specifically. They also found that RTs on go trials following threatening images were positively associated with self - reported social 4 anxiety. Similarly, i n an eye - tracking study, Buckner, Maner, and Schmidt ( 2010 ) found that those with high social anxiety took more time to disengag e fixations from disgust faces than those with low social anxiety. Together, these findings suggest that socially anxious individuals experience impaired cognitive control following the processing of threatening face s . Specifically, once attention is drawn to threatening faces, it is difficult for socially anxious individuals to disengage attention from these task - irrelevant stimuli and employ goal - directed attention to the task - relevant stimuli. However, in one study that implemented the dot probe task to investigate this relationship by Klump and Amir (2009), there was no evidence of delayed disengagement in social anxiety . A potential reason for this finding is that RTs in the dot prob e task do not lend themselves to disentangling rapid cognitive processes (i.e., isolating early attention from delayed disengagement). As such, more temporally precise measures are needed to further separate such rapid ly unfolding process es . A mple additional evidence suggests anxious individuals have impaired cognitive control ( Eyesenk et al., 2007; Moran et al., 2015; Krugg & Carter 2010) . However, f indings have been limited for social anxiety , in particular . Amir and Bomyea (2011) found that working memory performance on an operational span task (with threatening and neutral words) was impaired in those with SAD. Specifically , those with SAD were able to easily remember threat - relat ed words (i.e., exhibiting an attentional bias for threatening information), but they were not able to also remember neutral words (i.e., reduced cognitive control ). The researchers interpreted this finding as an indication that those with SAD experience a n inability to efficiently use cognitive control to inhibit the salience of threatening words and also remember neutral words . Moriya and Tanno (2008) found that higher levels of self - reported fear of negative evaluation (a core feature of SAD) , was associated with lower levels of self - reported attentional control on the Effortful 5 Control Scale (EC; Rothbart, A hadi, & Evans, 2000) . The EC is comprised of three subscales including attentional control, activated control and inhibitory control. I nterestingly , attentional control , defined as the ability to inhibit distraction, was the only subscale negatively correlated with social anxiety, while the other subscales were correlated with depression. T his relationship was consistent even when controlling for state anxiety and depression , indicating that SAD specifically contributes to attentional control impairme nts . Consonant with this finding , Weiser, Pauli, and Muhlberger (2009) found that individuals high in social anxiety had more antisaccade errors on trials including facial expressions , regardless of valence, suggesting a broad cognitive control impairment. In addition, in a non - emotional anti - saccade task, Liang (2018) found that those with social anxiety had longer latencies than those with low social anxiety . Liang (2018) concluded that this indicated a general inhibition impairment in individuals with so cial anxiety , that are not specific to emotional st i muli . Taken together, these findings imply that socially anxious individuals may experience broad cognitive control impairments across emotional and non - emotional stimuli. Event - related potentials (ERPs) as markers of attention b ias and cognitive c ontrol Because attentional bias and cognitive control processes are rapidly occurring functions, using reaction time measures do not allow for a precise measurement of their manifestation. Similarly, man y studies have noted the poor reliability of RTs in the dot probe task for assessing bias ( Schmukle, 2005 ; Brown et al., 2014 ; Macleod et al., 201 9 ) . For this reason, attempts have been made to analyze RTs in ways that can examine trial by trial changes throughout the task ( Zvielli, Bernstein, and Koster, 2014) , however this remains a distal proxy that can only measure downstream proces sing . Neuroimaging studies have aimed to address this by examin ing neural activity during the dot probe task (Monk et al ., 2006 ; Telzer et al., 2008; Price et al., 2014 ), 6 however they also s uffer from an inability to measure the precise time course of such dynamic processes. E vent - related potentials (ERPs) , on the other hand, serve as an excellent measure of online cognitive processing with millisecond precision. Importantly, because ERPs are temporally precise, it is possible to assess early attentional biases and cognitive control to the presentation of both emotional faces and the probe . Doing so will enable me to assess the time course of early attention al mechanisms (bias towards and delayed disengagement from threat) and later cognitive control durin g emotional face and the task - relevant probe processing . T he P1 and N2 are two such ERPs that index attentional and cognitive control mechanisms , respectively , and will be the focus of the present investigation . The P1 is an early positive deflecting ERP component that peaks between 70 and 150 ms post - stimulus onset. The P1 is a rapidly occurring signal generated from the parieto - occipital visual system (Luck & Kapperman, 2011) and its enlarged amplitude to threatening stimuli in social anxiety can be thought to index early negative attention bias. It has been shown to be sensitive to emotional faces ( Batty & Taylor, 2003) and modulated by amygdala damage (Rothstein et al., 2010) such that damage to this region resulted in a reduced P1 . Therefore, the P 1 is not only a marker for visual salience but is also influenced by the emotional salience of information. Within social anxiety, the P1 is enhanced for emotional faces, ( Kolassa & Miltner, 2006; Rossignol et al., 2012), threat - neutral face pairs in the dot probe task ( Santesso et al., 2008 ; Muelle r et al., 2009 ) , and positively correlated with social anxiety symptoms (Kolassa & Miltner, 2006) . Findings for the P1 to the probe in the dot probe task are limited , however , Mueller et al. (2009) observed that socially an xious individuals demonstrate a smaller P1 to probes replacing threatening faces . This finding is difficult to reconcile with extant studies showing speeded RTs on congruent trials. However, the P1 occurs earlier in time than the 7 execution of a response . As such, it could be that P1 ac tivity is reduced because attention al resources continue to be devoted to processing the previously presented threatening faces . This supports the notion that socially anxious individuals experience delayed disengagement from threatening stimuli . Therefore, it is also vital to assess change s in the P1 to the probe across treatment . The N2 is a negative - going signal that peaks approximat ely 200 - 3 5 0 ms post stimulus onset. It is tho ught to index recruitment of cognitive control and has been closely linked to inhibition processes in the presence of conflict (Folstein et al., 2008) . There is evidence that the N2 is generated from frontal cortical regions implicated in cognitive control , such as the DLPFC and ACC (Grossheinrich et al., 2013 ; Lavric, Pizzagalli, & Forstmeier, 2004) . Although m ost studies have examined the N2 in G o/ N o - go and the Flanker paradigms, past studies have also measured the N2 in the dot probe paradigm (Eldar & Bar - Haim, 2012; Thai et al., 2016). For individuals with social anxiety specifically, Thai et al. (2016) found that an attentional bias (measured via RTs) toward threat in socially anxious children was associated with reduced N2 amplitudes to faces in the dot probe task . This is consistent with the notion t hat biased attention is often coupled with a reduced implementation of cognitive control within anxiety . Therefore, the N2 is a prime candidate for examin ing changes in cognitive control across treatment for SAD . Additionally, due to the aforementioned research that cognitive control impairments are not specific to emotional faces, it will also be useful to assess cognitive control mechanisms to emotional and non - emotional stimuli i n the dot probe tas k i.e., fa ces and probes, respectively . Treatments for SAD : e ffects on a ttention b ias and c ognitive c ontrol Treatments aimed at relieving SAD symptoms include Cognitive Behavioral Therapy ( CBT ) and Attention Bias Modification Training (ABM) . CBT target s cognitive biases 8 surrounding social situations . It also promotes re - entry into social situations that individuals with SAD typically avoid (Huppert et al., 2003 ; Carpenter, Curtiss & Hoffmann, 2017 ). Individuals with SAD are asked to complete tasks, such as in vivo exposures ( e.g., engaging in a conversation with a confederat e ) and imaginal exposure s ( e.g., focusing on the worst possible outcome of social situations) that help with initiating, maintaining and ending social interactions (Huppert, Roth & Foa, 2003; Carpenter, Curtiss, & Hofmann, 2017) . Throughout treatment, individuals are instructed to focus on the conversations they are having , and widen their attentional window, instead of focusing on themselves (Huppert et al., 2003 ) or emotional faces that will confirm their fear s . Generally, these techniques produce significant reductions in both self - reported and clinician assessed symptoms ( Hein richs & Hoffman, 2005; Heimberg, 2002 ; Stangier, 2016 ; Huppert et al., in press ). T he re is also evidence that CBT aids in the reduction of attentional bias . In a review, Tobon et al. (2010) found that attentional biases in anxious individuals were reduced across a range of tasks (such as the dot probe and emotional Stoop) after CBT . However, f or SAD specifically, findings are limited and mixed. Mattia, Heimberg, and Hope (1993) found that CBT resulted i n a reduction of early attentional biases to social threat words using the Stroop paradigm, whereas Lundh and Öst (2001) did not find this effect using the same task . Additionally, neuroimaging studies indicate that amygdala activity is significantly reduced following CBT (Klump et al., 2013; Månsson et al., 2013 ) . To my knowledge, no study has examined P1 changes in the dot probe task throughout the course of CBT for those with SAD. There is also evidence to suggest that CBT increases engag ement of top - down corti cal brain regions implicated in cognitive control across a range of anxiety disorders ( Bruhle et al., 2014; P orto et al. 2009). Increases in frontal control regions following CBT are attributed to the 9 practice of reappraisal of negative situations that draw on and exercise frontal control regions . For in stance, Goldin et al. (201 3 ) found increased activity in the dorsolat eral prefrontal cortex (DLPFC) following CBT when those with SAD were asked to reapprais e negative self - beliefs . Similar findings were reported by Goldin et al. (2014), such that CBT increased recruitment of frontal cortical brain regions when SAD patients were asked to reappraise socially evaluative video scenes. No studies have examined changes in the N2 , as a marker of cognitive control, in patients with SAD following a course of CB T . As compared to CBT, ABM is a n intervention that specifically targets attentional biases by training attention away from threat (MacLeod, Rutherford, Campbell, Ebsworthy & Holker, 2002) . ABM follows the parameters of the dot probe paradigm, with the exception that probe s replace neutral face s in most or all of the trials that is, attention is trained to focus on neutral faces instead of negative faces because that face predicts the location of the target most of the time . One advantage of ABM is tha t it is easily scalable and requires little effort from the patient to engag e in it. ABM has shown positive effects on anxiety symptoms ( Amir, Weber, Beard, Bomyea, & Taylor, 2009; Mogoase et al., 2014 ) . However, a recent meta - analys i s by Mogg, Waters, Bra dley (2017) suggests the effects of ABM are more modest than previously documented . Similarly , r esults for effect on attentional bias es ha ve been mixed ( Heeren, . There is some evidence that ABM reduce s attentional biases (Heeren, Lievens, & Phil ippot, 2011 ; Schmidt, Richey, Buckner, & Timpano, 2009) , but t here are also studies that have found no improvements (Bunnell, Beidel, & Mesa, 2013 ; Julian, Beard, Schmidt, Powers, & Smits, 2012 ) . T hese mixed findings highlight the need for additional measures of bias change in the dot probe during ABM, such as ERPs. Evidence from past ERP research of ABM fo r anxiety is limited , however , 10 found a reduced P1 to face pairs in the dot probe task after ABM for trait anxious individuals trained away from threat . T o my knowledge , no study has investigated changes in the P1 in the dot probe task after ABM for individuals with SAD . T here is also evidence that ABM improves cognitive control. S upport comes from a study conducted by Eldar & Bar - Haim (2010) who found that individuals high in trait anxiety showed enhanced N2 amplitudes to face pairs in the dot probe task after ABM training . Dennis et al., (2017) found increased N2 amplitudes to threatening faces in the dot probe after a mobile delivered version of ABM training. T herefore, there is precedence for enhanced N2 to face pairs following ABM , which suggests that ABM improves the recruitment of cognitive control during emotional face processing . Notabl y , these studies did not a ssess changes in the N2 elicited by the probe. Examining the N2 to the probe will allow for an assessment of cognitive control to task - relevant stimuli, which I plan to do in the current study. Importantly, although prior research points to the effects of CBT and ABM on attention bias and cognitive control , no study has assessed changes in both mechanisms in a comparative context of CBT and ABM for SAD. A recent report using behavioral and symptom data from the same sample to be used for the current study found that attentional biases measured via RT s did not change across CBT and ABM treatments, although there were significant improvement in symptoms (Huppert et al., in press ). This null effect on RTs highlights the need for other measures to assess change in attentional biases and cognitive control across treatment , which serves as the central motivation of the present study . The Present Study In sum, little research has focused on neurophysiological change across two active treatments for SAD . Thus, the present study has a primary aim of compar ing the modulation of 11 ERPs indexing early attention bias and cognitive control by ABM and CBT treatments. The P1 will be used to assess changes in early attention bias , and the N2 will be used to assess changes in cognitive control. These waveforms will be examined to both the faces and the probe in the dot probe paradigm to disentangle cognitive processing dynamics over the course of a trial . Using ERPs in this way will allow for an investigation of both change s in early attentional allocation to and cognitive control during emotional face processing as well as changes in the processing of task - relevant stimuli (i.e., the probe ) . Furthermore, this study has a secondary goal of examining whether neurophysiological change, as measured by the P1 and N2, relate to symptom change . To my knowledge, no extant studies have attempted to link ERP changes in attentional bias and cognitive control with symptom improvement in two active treatment for SAD. There are five hypotheses to address the first aim. During face processing , I predict that (1) the P1 amplitude will decrease over treatment , index ing less attentional bias to the faces, and (2) the N2 amplitude will increase over treatment , indexing more cognitive control during face processing . For the probe, I predict that (3) the P1 amplitude will increase on congruent trials over treatment indic ating enhanced attention to the probe (i.e., disengagement from threatening faces), and that (4) the N2 amplitude will increase over treatment, indicating more goal - directed cognitive control to the probes. Lastly, I predict that (5) these ERP changes will be strongest in the ABM group. I predict that ABM will show enhanced effects for two reasons. Conceptually, ABM specifically targets cognitive mechanisms in the dot probe task whereas CBT does not, thus effects on neurophysiological indices of cognition e licited in the dot probe task should be larger for ABM. Second, there is more empirical data to support that these specific neurophysiological changes will be evident in the dot probe following ABM, and no such 12 supportive data for CBT. With regard to the second aim, there are two hypotheses. I predict that ( 1 ) decreased P1 and increased N2 activity to faces will relate to symptom reduction . I further predict that (2) increased P1 and N2 activity to the probes will relate to symptom improveme nt . 13 METHOD Participants Descriptive s tatistics for self - reported age, gender, and handedness are reported in Table 1. Part icipants were recruited from advertisements or referrals from doctors and clinics in Israel . Criteria included participants who were 18 years or older and met eligibility after screening for social anxiety. Social anxiety was assessed using the Mini International Neuropsychiatric Interview (MINI; Sheehan, 2006) and the Lebowitz Social Anxiety Scale (i.e., scores >50) (LSAS; Liebowitz, 1987) . Exclusionary criteria included any history of psychosis, bipolar disorder, suicidality, or current substance abuse . The fina l sample included a total of 5 0 people assigned to either CBT ( n = 3 3 ) or ABM ( n = 17 ) (see Figure 1) . Measures All measures were translated from English to Hebrew and back - translated by another individual for validity. Leibowitz Social Anxiety Scale (LSAS; Leibowitz, 1987). The LSAS is a semi - structured interview to screen participants for social anxiety symptom severity . It assess es for anxiety and avoidance in social and performance situations. It consists of 24 items that participants rate for level of fear from 0 ( Never ) to 3 ( Severe ), and avoidance from 0 ( Never ) to 3 ( Usually ). A global sum score ( across both fear and avoidance) was calculated to assess symptom severity for those with social anxiety. Higher global scores indicate more severe social anxiety . Th e interviewers were doctoral students in clinica l psychology who administered the LSAS in four - week intervals as part of monthly assessment. They also completed the LSAS at the end of treatment, and at a three - month follow up. They were blind to experimental conditions and had no other interactions with the participants . Inter - rater reliability was 14 conducted by videotaping 15 evaluations and a second person rated the LSAS while watching. I nter - rater reliability was high ( r = .94, p <.01 ). Another reliability test was conducted by randoml y selecting 15 evaluations to be administered a second time within the first week of intake assessment, which also produced high reliability ( r = . 80 , p <.01 ). Social Phobia Inventory (SPIN; Conner et al., 2000) . The SPIN is a 17 - item questionnaire that asks participants to rate their level of fear, avoidance, and arousal in the past week. The SPIN was shown to have reliable psychometric properties in clinical populations (Connor et al., 2000) . Participants are asked to respond to items using a 5 - poi nt Likert scale ranging from 0 ( Not at all ) to 4 ( Extremely ). All scores are summed to create a total score for the questionnaire, with higher scores indicating more social anxiety. Participants completed the SPIN in their initial assessment, and also completed it as part of their weekly assessments to track for self - reported symptom change throughout treatment. The SPIN was also given at the end of treatment, as well as the three - month follow - up. Dot Probe Task . In this task , trials begin with the pre sentation of a fixation point at the center of the screen for 500 ms. Then, two images of fac ial expressions were presented vertically one above and one below the fixation cross . A ll trials included one neutral and one threatening (i.e., angry, disgust) facial expression. Both images terminated after 500 ms and one of the images was replaced by an arrow (probe). Participants were required to respond to the direction of the arrow with their dominant hand on a response box. A total of 64 facial expression images were selected from NIMSTIM (Tottenham, Borscheid, Ellertsen, Marcus, & Nelson, 2002) and TAU (Frenkel & Bar - Haim, 2006) databases , with threatening images including anger, disgust or contempt. The task consisted of 256 trials in which all threat - neutral pairs were presented four times , with probes replacing the threatening 15 face 50% of the time . Therefore, half of the trials were con g ruent (probe replace d a threatening face), and half were incongruent (probe replace d a neutral face). This task was administered at the first assessmen t and every four sessions during treatment. Participants also completed this task at the end of treatment and three - month follow up. Electroencephalography (EEG). Continuous EEG activity was recorded from 64 Ag - AgCl electrodes fitting into a BioSemi (BioSemi, Amsterdam, The Netherlands) stretch - Lyrca cap while participants completed the dot probe task . The cap consists of electrodes placed across the scalp to cover locations across the cerebral cortex, including central, frontal, parietal, temporal, and occipital lobes. The ca p size was chosen by measurin g the circumference of the head. Once the appropriately fitted cap was placed on the head, measurements were taken to ensure that electrode channels were placed over the appropriate region of the brain . These measurements were taken from the nasion (a prom inent bump on the front of the head, usually between the eyebrows) to the inion (a prominent bump in the back of the head indicating the end of the skull). Additional measurements were taken from the top of the left ear to the right ear. The cap was fitted with a Velcro chinstrap to ensure its stability during data acquisition. Two additional electrodes were placed on the right and left mastoid bones , a relatively prominent bone behind the ear, to serve as a reference. E lectro - oculogram (EOG) activit y generated by eye movements and blinks was recorded at FP1 and by electrodes placed above and below the left eye, and on the right and left of the outer canthi (the outer corner of the eye). During data acquisition, scalp recordings were referenced online t o two electrodes called the common mode sense (CMS) active electrode and the drive n right leg (DRL) passive electrode. CMS serv e s as a ground electrode and DRL is part of a feedback loop that drives the zero . Further analyses were conducted offline. 16 EEG data was processed offline using Brain Vision Analyzer 2 (Bra i n Products, Gilching, Germany). The recordings were re - referenced to the numeric mean of the right and left mastoid bones , such that the activity re corded at the mastoid bones was subtracted from the activity recorded from scalp electrodes to isolate activity on the scalp. Next, activity on the scalp was band - pass filtered to include activity in the frequency band between 0.1 30 Hz (12 dB/oct rollo ff). Blinks and eye movements were corrected using the Gratton, Coles, & Donchin (1983) method. This method corrects for the confounding effects of eye movements and blinks using a regression - based method. This is achieved by averaging across EOG and scalp recorded activity and estimating propagation factors by regressing average EOG activity onto scalp activity and correcting EEG data based on the estimated propagation factors. Trials where errors were committed were removed, and all correct trials were segmented into epochs starting 100 ms pre - stimulus onset and continued for 600 ms post - stimulus onset. This was completed twice: for the onset of the faces and the onset of the probes. Trials were rejected based on the following criteria: voltage step betw een two contiguous data points exceeding of 50 microvolts (uV), a voltage difference of more than 200 uV within a trial, or a maximum voltage difference less than .5 millivolts within a trial. All of these occurrences do not reflect natural occurring neuro physiological activity but rather poor connection to the scalp, sudden head movements and, and other related artifacts. Upon visual inspection, the P1 was extracted in the 80 - 150 ms time window. T he sites were p ooled at parietal occipital sites on the left ( O1, P3, P5, P7, PO3, PO7 ) and right ( O2, P4, P6,P8, PO4, PO8 ) of the head. Because the faces are vertically and centrally presented, I did not hypothesize that the average activity in the P1 would differ as a function of threatening face location. Therefore, the activity across the specified time window was averaged across face 17 location (top and bottom) and congruency (probe replacing threatening and neutral faces). Visual inspection revealed a clear difference in activity on the left and right of t he head, however. Because of this, an initial model was tested to examine if lateralized location (left vs. right of the head) interacted with time or group. The model revealed a main effect of lateralized location, such that the right side of the head was more positive than the left ( b = .24, t (550) = 2.199, p = .03) (see Figure 2 and Results below) 1 . However, this did not interact with time, or group nor was there a three - way interaction (all p >.11). Therefore, the final model averaged across the left and right of the head. Next, the N2 was identified as the average activity in the 200 - 350 ms time window at site Fz . Additionally, models for site specification to the probe were first tested with a fixed effect for time, congruency, and group. This was done to examine whether congruency interacted with any additional terms. There was no effect of congruency nor did cong ruency interact with time or group (all p - locked to probe presentation were averaged across location and congruency. Visual inspection for the probe - locked P1 revealed there was no early positive activity to the probe, contrary to my hypothesis. However, there was bilateral negative activity at parietal - occipital sites, indicating there was an N1. Therefore, the average activity between 70 and 140 ms was pooled across lateralized sites on the left (P7, PO7, O1 , PO3, P5, P3, P1) and right of the head (P2, P4, PO4, P6, PO8, O2, P8) 2 . Upon visual inspection, the N2 was specified as the average activity between the 215 - 350 time window at site Fz . 1 A model was tested with an additional interaction terms for lateralized effect, including location X Group, location X time, and location X follow - up. These did not reach significance (all p = .32) , indicating that it also did not interact with follow - up. 2 To remain consistent with the face - locked analyses, a lateralized effect was tested. The model revealed small estimates for time ( b = - 2.449 e - 17) and time X lateralized location ( b = 7.6883 e - 1 8), causing a large p - value ( p =1). Therefore, this effect was excluded from the final model. 1 8 Treatments Cognitive Behavioral Therapy. Patients underwent up to 2 0 sessions of CBT on a weekly basis . Sessions lasted approximately 60 - 90 minutes per week. CBT was delivered by graduate students under the supervision of Dr. Jonathan Huppert , a doctoral - level , licensed expert clinician. T herapy followed the protocol s of Huppert, Roth Ledley, and Foa (2006) and of Clark (2005) . It s main components included the manipulation of self - focused attention and safety behaviors, strategies that those with SAD perform when confronted with anxious situations ( e.g. , avoiding eye contact) . For manipulation of self - focused attention, patients practic ed shift ing attentional focus between internal stimuli (self - focused thoughts, feelings, images) and external stimuli (individuals with whom one interacts, facial expression of others, colors and shapes of buildings, etc. ). Patients are instructed to resist safety behaviors. A main component of therapy was exposure exercises (i.e., behavioral experiments ) in which participants were instructed to engage in fearful social situations this was done in session and out of sessio n for homework . The treatment also incorporated video and audio feedback of the patients during their exposures exercises , enabling them to see themselves and an observer, and the interaction as a whole. B uilding on treatment techniques by Huppert et al. (2003) , treatment also incorporated developing an idiosyncratic model linking negative beliefs about social interactions to internal atten tional focus, safety behaviors, physiological symptoms and overt avoidance of social situations. I maginal exposures to social situations and social skills training were also included . The re was no formal coding of the adherence to therapy. Attention Bias Modification (ABM). Up to eight sessions of ABM was delivered on a weekly basis. Sessions las t ed approximately 10 - 15 minutes per week. The ABM procedure followed a modified d ot p robe protocol. A fixation cross was presented on the screen for 60 ms 19 followe d by the presentation of threat - neutral face pairs at the center of the screen for 500 ms. To train attention away from threat, the probe replaced the neutral face on 80% of the trials . Procedure Participants were screened in person for clinical disorders and SAD using the MINI, LSAS, and a reaction time task 3 . Following this, all participants completed an assessment consisting of the dot probe task, the SPIN and the LSAS. Then, p articipants were randomly assigned to either the CBT or ABM condition and completed their re spective treatment. Four - week assessments were administered in which participants completed the dot probe task, SPIN , and LSAS. Participants also completed an assessment at the end of treatment (up to 8 weeks for AMB and up to 20 weeks for CBT), and three - month follow - up (see Figure 2 ) . Data Analytic Approach The & Walker, 2015) ( Pinheiro, Bates, DebRoy, Sarkar, & R Core Team, 2016) packages in R Version 3.5 . 1 . The first aim was analyzed using multi - level models (MLMs) . The analyses took an intent - to - treat approach, with all subjects having at least one observation after the pre - treatment assessment , as MLM can estimate effects with missing observations . For the first aim, I conducted a series of growth models for each ERP. Before conducting each model, I computed an interclass correlation (ICC), which c omputes the amount of variability contained between people, which provides support for a random intercept model. All models were fit using 3 1 P articipants also had to meet a threshold of 75% accuracy on a Grammatical Decision Task (GDT) to be eligible to participate . 20 restricted maximum l ikelihood estimation. Model fit was assessed by computing the proportion of variance explained at level 1 (i.e., time) and level 2 (group) between the unconditional and random slope models. For all models, group (effect coded) was entered as a cross - level interaction with time and follow - up to examine if change in ERPs differed between treatment groups. Time was centered with the pre - treatment assessment as baseline, with a one unit increase in time being equivalent to the passing of approximately one month . Two models were estimated to examine rates of change (1) at the 8 - week mark (i.e., post - treatment for the ABM group) and (2) at post - treatment for both groups (i.e., up to 20 weeks for the CBT group). For the first model, only observations up until the 8 - week mark were included. This resulted in the following model - Level 1 Model: Level 2 Model: Combin ed Model: For the second model (estimating the rates of change at the post - treatment point for both groups), all observations were used with a random intercept and slope. Included in these analyses (i.e., 21 those examining post - treatment change) was a test of follow - u p (dummy coded), which was done by fitting a piecewise model. This allowed for the examination of change from post - treatment to follow - up. This resulted in the following model Level 1 Model: Level 2 Model: Combined Model: In sum, all models had three primary effects of interest a time effect, which would indicate a monthly ERP change rate over time; a group effect which would indicate different levels of average activity between groups averaged across time, and a time X g roup cross - level interaction, which would indicate different rates or direction of change between groups. Lastly, in models analyzing post - treatment change, it was of interest to examine if there were differences in ERPs at the follow - up assessment, and wh ether these effects differed by treatment group. Next, two approaches were taken to evaluate the second aim whether changes in ERPs relate to symptom change. The first approach involved conducting separate growth models (until 22 post - treatment) predicting ERPs (P1/N2 to the faces and P1/N2 to the faces) and symptoms (LSAS and SPIN scores) with a random intercept and slope for time using restricted maximum bivariate correlations were conducted to examine if changes in ERPs related to changes in symptoms. This set of analyses allowed for the investigation of the relationship between the rates of change in ERPs and symptoms throughout the course of treatment. The secon d approach involved computing multilevel models with lagged pre - treatment centered ERP values as the predictor of symptoms. In these models, I included an autoregressive covariance structure for time, simply meaning that the covariance structure controlled for the fact that ERP values closer together in time are more correlated than values further apart. These models allowed for a more fine - gained analyses of ERP and symptom relationships, in that they examine d whether the ERP amplitude measured at one time - point predicted symptom levels at the following time - point. These models also included a group effect (effect coded) to examine if changes in symptoms differed by group, as well as a group by ERP interaction to assess if the lagged relationship between ER Ps and symptoms differed by group. For these models, model fit was assessed by comparing the amount of variance contained in the final model and the amount of variance in the dependent variable (obtained in the unconditional model with no predictors). 23 R ESULTS Face - locked Growth Models I first examined the rate of change at the 8 - week mark of treatment for the P1 to the faces . The unconditional model revealed an ICC of 65%. The final model included a random intercept and slope for time and revealed that there was no effect of time ( b = .02, t (179) = .205, p = .84), group ( b = - .46, t (179) = - 1.52, p = .14), or group X time interaction ( b = - .07, t (179) = - .75, p = .46). The second model estimated the rate of change of the P1 to the faces until the post - t reatment assessment for both groups. The unconditional model revealed an ICC of 60%. The final model revealed no main effect of time ( b = - .05, t (275) = - .533, p = .60), group ( b = - .07, t (275) = - .162, p = .87), or group X time interaction ( b = - .13, t (275) = - 1.318, p = .19). The results also showed there was no follow - up effect ( b = - .17, t (275) = - 1.32, p = .19), or group X follow - up interaction ( b = - .34, t (275) = 1.398, p = .16). The results imply that treatment did not affect early attentional pro cessing to threatening faces, nor did this differ by groups. They also imply that the early attentional processing to threatening faces, as measured by the task, remained unchanged even when treatment ceased. Next, the N2 to the presentation of faces was e xamined at the 8 - week mark. The unconditional model revealed an ICC of 64%. The final model included a random intercept and slope for time. The model revealed a main effect of time, such that the N2 increased across both treatments, ( b = - . 41 , t( 179)= - 2.605, p =.0 1). There was no group effect ( b = - . 22 , t (179 )= - .508, p =. 61), nor was there a group X time interaction ( b = - . 10 , t( 179)= - .63, p =. 53) (see Table 2). This model revealed that proportion of variance explained at level 1 (i.e., time) for th is model was 11% more than the unconditional model . 24 The unconditional model for the N2 to the faces with all observations revealed an ICC of 52%. The effect of time remained , such that the N2 increased over time for both groups ( b = - . 38 , t( 275)= - 2.3, p =.0 2) (see Figure 3) . This was not qualified by a group X time interaction ( b = - . 11 , t ( 275)= - .67, p = .50) nor a group effect ( b = - . 22 , t( 275)= - .475, p = .63) . There was no change at f ollow - up ( b = . 70 t( 275)= .81, p = . 42) no r a group X f ollow - up interaction ( b =. 32 , t( 275)= - .365, p = . 72), indicating that changes in the N2 were sustained at follow - up for both groups (see Figure 4 and Table 3). This model revealed that 12.4% of the variance was explained with time (in comparison to the unconditiona l model). Together, the ERPs time - locked to face presentation evidenced no change in the P1 . However, there was a significant increase in the N2 over time , seen both at the 8 - week mark and at the end of treatment for both groups. Probe - locked Growth Mode ls First, the effect of time on the N1 was examined at the 8 - week mark. The unconditional model revealed there was 23% of variability between groups. The results of the final model revealed that there was no effect of time ( b = - .16 , t( 164) = - 1.26, p = . 21), group ( b = .41 , t( 164)= 1.58, p = . 11), nor an interaction ( b = - .11 , t( 164)= - .84, p = . 41). For the model with all observations used, the ICC was 28%. This model showed there was no effect of time ( b = .09 , t( 247)= .86, p = . 39). There was a marginal effect of group ( b = - .46 , t( 247)= 1.85, p = . 07), such that the ABM group had a larger N1 than the CBT group. This implies that overall, the ABM group evidenced more attentional capture to the probe. This effect was not qualified by a group X ti me interacti on ( b = - .17 , t( 247)= - 1.55, p = . 12). There was also no effect of follow - up, nor a group by follow - up interaction (all p At the 8 - week mark for the N2 , the ICC for the model revealed was 28%. The results showed there was no effect of time ( b = - .27, t (175) = - 1.12, p = .26), group ( b = - .14, t (175) = - 25 .32, p = .75), or group X time interaction ( b = - .01, t (175) = - .09, p = .93). The final model including all observations revealed a marginal effect of time, such that the N2 increased over time ( b = - .33, t (271) = - 1.709, p = .09). There was no effect of group ( b = - .14, t (271) = - .32, p = .75), or time X group interaction ( b = - .05, t (271) = - .23, p = .82). There was also no effect of follow - up ( b = 1.40, t (271) = 1.41, p = .15), nor was there a follow - up X group interaction ( b = - .1, t (271) = - .10, p = .92). In sum, there was no change in the N1 to the probe over time , however the ABM group evidenced more attentional capture (i.e., a larger N1) to the probes overall. Lastly, there was a marginal change in the N2 over time for both treatment groups , such that the N2 increased to probe presentation. Relationship betwee n ERP change and symptom change First, individual growth models were computed for each ERP and for symptoms. These models only included the post - treatment mark, with a random intercept and slope for time (see Tables 4 - 11). This allowed for the extraction of the slopes for each individual, which were used for bivariate correlations. There were no relationship s between ERPs an d symptoms (all p (see Table 12). However, there were relationships among ERPs and among symptoms 4 . The P1 evidenced a positive relationship with the N2 to the faces ( r (47) = .32, p < .02) and the N1 to the probe ( r (45) = .3 0 , p = .04) (see Figur e 6 - 7 ). These results imply that the higher the rate of change in early attention processing to the presentation of faces, the smaller the rate of change in cognitive control to the faces and attentional capture to the probe , respectively. The N2 to the fa ces evidenced a marginal positive relationship to the N1 to the probe ( r (45) = .26, p < .08). 4 An outlier was identified fo r these sets of analyses. Upon removal, previously significant relationships were no longer reliable. Therefore, the results reported here exclude one participant. 26 This implies that the smaller the rate of change of cognitive control to the faces, the smaller the attentional capture to the probe. The re were also high positiv e relationships between symptom changes (all p suggesting that the decrease in symptom s over time related to one another across all measures (see Table 12). Multilevel Models with Lagged Pre - Treatment Centered ERPs to Face Presentation as Predictors Next, the lagged models were conducted to obtain a more fine - grained analyses of the relationship s between the ERPs and symptoms. For the P1 to the face, the models showed there was no effect of the P1 on LSAS total scores ( b = - .20, t (182) = - .26, p = .79), nor was there a group effect ( b = - 2.86, t (182) = - .88, p = .38), or group X P1 interaction ( b = - .87, t (182) = - 1.13, p = .26). Similar effects were found with LSAS A nxiety (all p subscale scores (all p 33). For the SPIN, there was also no effect of the P1 to face presentation, ( b = .31, t (182) = .63, p = .53) ; however, there was a marginal group effect ( b = - 3.28, t (182) = - 1.9, p = .06), such that the CBT group had lower estimates ( b = 32.66) than the A BM group ( b = 39.16). There was no group X P1 interaction ( b = - .24, t (182) = - .48, p = .64). Therefore, there was no relationship between early attention to face presentation and symptom measures. For the N2 to face presentatio n, the N2 did not predict LSAS total scores ( b = - 53, t (182) = - .95, p = .35) nor was there a group effect ( b = - 2.76, t (182) = - .84, p = .40). There was , however, a group X N2 interaction ( b = 1.17, t (182) = 2.07, p = .04). Simple slopes analyses showed that there was a difference in the direction of the slopes, but neither of the slopes reached significance for the ABM ( b = - 1.71, t (182) = - 1.64, p = .10) or CBT group ( b = .64, t (182) = 1.46, p = .15). Investigation of the anxiety and avoidance subscale scores on the LSAS 27 illumina ted these findings , however . The results revealed that the above effect is mostly influenced by avoidance scores . Specifically, the final model showed there was no effect of N2 ( b = - .44, t (182) = - 1.45, p = .15), nor was there a group effect ( b = - 1.36, t (182) = - .75, p = .46), but there was a significant N2 X group interaction ( b = .78, t (182) = 2.56, p = .01). This model explained 2.1% additional variance over the unconditional model. Simple slope analyses revealed that the ABM group evidenced a negative relationship between N2 and LSAS Avoidance ( b = - 1.23, t (182) = - 2.17, p = .02), such that a smaller N2 (i.e., less negative) predicted less symptom s at the next time - point. There was no relationship for the CBT group ( b = .34, t (182) = 1.45, p = .15) (see Figure 8 ). There was no effect of the N2 on LSAS Anxiety scores, nor was there a group effect or group X N2 interaction (all p Additionally, relationships between ERPs and SPIN scores also showed effects that marginally differed by t reatment condition . There was no effect of the N2 ( b = .15, t (182) = .44, p = .66). There was a marginal group effect ( b = - 3.11, t (182) = - 1.90, p = .06), such that the CBT group had lower estimated scores ( b = 32. 75) than the ABM group ( b = 40.06). However , there was also a marginal group X time interaction ( b = .58, t (182) = 1.71, p = .08). Simple slope analyses showed that there was a positive relationship between the N2 and SPIN scores for the CBT group ( b = .72, t (182) = 2.61, p = .01), but no re lationship for the ABM group ( b = - .43, t (182) = - .69, p = .49) (see Figure 9) . That is , a larger N2 at one tim e - point predicted lower symptoms at the next time - point. This model explained 17.24% additional variance over the unconditional model. Overall, these results imply t hat there was no relationship between the P1 and symptoms and that this did not differ by treatment group. H owever there was the relationship between the N2 and LSAS A voidance for the ABM group only, such that a smaller (i.e., less ne gative N2) predicted higher symptoms throughout the course of treatment . On the other hand 28 there was a marginal difference between the N2 and SPIN, such that a larger N2 predicted less symptoms for the CBT group only. Taken together, the results of the la gged models revealed no relationship between the P1 time - locked to face presentation and symptoms, suggesting early attentional bias to threatening faces were not related to symptoms. However, there was a significant relationship between the N2 time - locked to the faces that differed by treatment group. Specifically, a smaller N2 at one time - point predicted lower LSAS A voidance symptoms at the next time - point for the ABM group only . On the other hand, the relationship between the N2 and symptoms in SPIN scores marginally differed between groups , such that a larger N2 at one time - point predicted less symptoms at the following time - point for the CBT group . Multilevel Models with Lagged Pre - Treatment Centered ERPs to Probe Presen tation as Predictors The N1 to the probe showed no effect on LSAS total scores ( b = - 1.17, t (171) = - 1.18, p = .24), no group effect ( b = - 2.13, t (171) = - .65, p = .52), and no group X N1 interaction ( b = 1.31, t (171) =1.32, p = .19). Similar effects were found for LSAS A nxiety scores (all p Avoidance scores showed a slightly different pattern, however. Although there was no effect of the N1 ( b = - .81, t (171) = - 1.52, p = .13) or group ( b = - .87, t (171) = - .49, p = .62), there was a marginal group X N1 interaction ( b = .94, t (171) =1.77, p = .08). The follow - up with simple slopes revealed that none of these results reached significance, but both groups showed opposite relationships with LSAS Avoidance scores. Spec ifically, the CBT group showed no relationship ( b = - .13, t (171) = - 49, p = .63), but a marginal negative relationship for the ABM group ( b = - 1.75, t (171) = - 1.7, p = .09) , suggesting that a smaller N1 (i.e., less attentional capture to the probe) predicted lower symptoms. There was no relationship between the N1 to the probe and 29 SPIN scores ( b = - .20, t (171) =.33, p = .74), nor was there a group X N1 interaction ( b = - .009, t (171) = - .01, p = .99). Consistent with the above models, there was a grou p effect ( b = - 3.46, t (171) = - 2.02, p = .04), with lower estimates for the CBT group ( b = 32. 54) than the ABM group ( b = 3 9.45). For the N2 to the probe, there was no effect of the N2 on LSAS total ( b = - .57, t (179) = - 1.01, p = .32), no group effect ( b = - 2.68, t (179) = - .83, p = .41) or group X N2 interaction ( b = - .28, t (179) = - .51, p = .61). Similar effects were found with LSAS Anxiety (all p Avoidance (all p scores ( b = .10, t (179) = .27, p = .78), a marginal group effect ( b = - 3.28, t (179) = - 1.96, p = .05), and no ( b = .25, t (179) = .69, p = .49). The group effect remained consistent, in that there were lower estimates for CBT ( b = 32. 56) than ABM ( b = 3 9 . 47). All in all , there was a marginal effect for the N1 for the ABM group, suggesting that less attentional capture to the probes was related to lower symptoms. Lastly, there was no relationship between the N2 to the probe and symptom measures, suggesting that cognitive control to the probe was not rel ated to symptom change across both treatment groups. 30 DISCUSSION Cognitive models of anxiety have indicated that individuals with social anxiety experience enhanced attention bias to ward threatening stimuli, such as faces, which dampens the ability to exercise top - down cognitive control. There is evidence that CBT and ABM can aid in reducing attentional bias and increasing cognitive control, however , it remains unclear if there are differ ences between each treatment to effectively do so. In addition, it remains unclear if indices of these cognitive processes within each treatment differentially relate to symptom reduction. I found that the N2 to the faces significantly increased across bot h CBT and ABM, while there were no changes in the P1 to the faces or the N1 and N2 to the probe. Additionally, the rate of change of these ERPs did not relate to the rate of change in symptoms reported in the LSAS or SPIN. Lastly, we found that N2 scores to the faces were negatively related to LSAS A voidance symptoms for the ABM group only , such that the smaller the N2 at one time - point, the higher the symptoms at the next time - point . On the other hand, there was a marginal group difference for the N2 and SPIN scores , such that a larger N2 predicted less symptoms for the CBT group. Overall, results from the current study provided mixed support for hypotheses. I nconsistent with my predictions , results showed that there was no change in the P1 amplitude to face presentation , indicating no change in attentional capture to faces across both treatments. The literature for the modulation of attentional bias for both CBT and ABM are limited and mixed (Mattia, Heimberg, & Hope, 1993; Lundh & Öst , 2001) . In addition, in a report using the data from this study, there was no change in behavioral metrics of attentional bias both the traditional calculation and trial level variability in either group (Huppert et al., in press ). There seem to be two possible explanations for these findings. First, because attention biases are fast 31 and automatic, they may just be difficult to change. Second, issues of reliability of early attention bias es have been raised (Schmukle, 2005; Brown et al., 2014; Macleod et al., 201 9 ) . That is, it may be difficult to detect change in an unreliable metric . To counter t his , the argument has been made that attentional bias may be better conceptualized as a probabilistic phenomenon that pre - disposes anxious individuals to be more likely to have an attentional bias toward threat. However, the stability and strength of the b ias depends heavily on the contextual factors that exacerbate it (MacLeod et al., 2019). This conceptualization would continue to mean that its measurement is unreliable (unless contexts are stable at every assessment). Therefore, p erhaps the processes exe rcised in CBT and ABM are not modulating these dynamic and context - dependent changes in early attentional processes, or at the very least, the change in early attention elicited by these treatments is quite difficult to capture. On the other hand , the N2 time - locked to face presentation increased as a function of both treatments. This signifies that both treatments exercise d the engagement of frontal brain processes associated with cognitive control . I n addition, the N2 increased at the 8 - week mark for bot h groups (which is approximately half way through treatment for the CBT group) , suggesting that the engagement of cognitive control processes occurs relatively early in treatment. These results are aligned with other studies that have found that the N2 inc reases as a function of ABM (Eldar & Bar - Haim, 2010), and CBT increases the engagement of frontal brain regions (i.e., the DLPFC) associated with cognitive control function (Goldin et al., 2014). As such, it can be concluded that the engagement of these to p - down processes in the presence of emotional stimuli may be a useful mechanism of change to target and measure throughout the course of treatment. Additionally, c ontrary to my hypothesis, these data did not evidence a P1 to the probe but did evidence an N1 . The N1 time - locked to targets in the dot probe task has been interpreted to 32 index facilitated attention (Torrence & Troup, 2016; Zhang et al., 2016) , and its interpretation is similar to that of the P1 waveform. There was no change in the N1 to the p robe over time, suggesting that neither group produced change in delayed disengagement. Similarly, Klump and Amir (2009) also did not find evidence for delayed disengagement using this task. These results further speak to the difficulty to measure delayed disengagement in the dot probe task. H owever, the ABM group evidenced a larger N1 overall . This could be due the repeated exposure that the ABM group ha d to the task parameters, both for assessment and treatment, that resulted in more facilitated attention to the probe. Partially supportive of my hypothesis was a marginally significant increase in the N2 over time to probe presentation. This implies that there was increased cognitive control to mitigate the interference of the salient face distractors. With the exception of a few studies (Mueller et a l ., 2009, Zhang et al., 2016) , the majority of the literature on the dot probe task has not examined change in ERPs to the probe , and no studies have examined change in the N2 . The results also speak to the need for increased measurement of the N2 to probe presentation. Doing this will allow for an examination of the ability to not only recruit cognitive control during emotionally salient information, but to also do so for a task - relevant goal (allowing for the ability to observe if treatment also helps facilitate increased ability to complete task demands) . This may provide a useful way for to examine how both treatments improve sustained goal maintenance , which is often impaired in individuals with anxiety . In sum, the change in ERPs produced by CBT and ABM broadly imply that both treatments serve to primarily exercise rapid proc esses influenced by frontal - cortical regions of the brain in the presence of both emotional and non - emotional stimuli. The fact that the ABM group the treatment aimed to specifically target attentional bias did not produce change in 33 attention al bias , could further imply that such early attentional processes are difficult to target, and therefore difficult to change. However, it may be more apt to identify CBT and ABM as treatments that increase the recruitment of frontal - cortical regions of the brain during the dot probe task for those diagnosed with SAD . Furthermore , the second aim sought to examine if changes in these ERPs were related to symptom change. Contrary to my hypothesis, although neither of the ERP change slopes were related to symptom ch ange slopes , ERP changes were related to one another, and symptom changes were related to one another. This implies that treatment produced change at a similar rate among ERPs and symptoms, but the rate of change in the neurophysiological index did not rel ate to the rate of change in reported symptoms of anxiety . These results speak to a broader discourse of the differential change seen in the brain and reported symptoms in individuals diagnosed with SAD . More research is needed to further elucidate these findings, such as examining if changes in ERPs are specific to the dot probe. Results of lagged models, however, showed that changes in the N2 had differential effects on symptoms across groups. Specifically, for the CBT group, changes in the N2 were unre lated to LSAS symptoms . On the other hand, for the ABM group, a larger N2 predicted higher LSAS A voidance symptoms at the next time - point, but no relation to LSAS A nxiety or SPIN symptoms. These findings are rather surprising because both groups evidenced increases in cognitive control to the faces, however, this increase seem ed to be harmful for the ABM group in the avoidance domain. Primarily, i t is important to note that cognitive control is a global term, that could be used to delineate a range of proce sses (e.g., top - down orienting, inhibitory control) during stimuli conflict such as what is presented in the dot probe task (Mogg & Braddley, 2017). All of these processes require the recruitment of frontal cortical regions. Because the faces in the 34 task o r centrally and vertically presented and there are no neutral - neutral face pairs, it is unclear whether the increase in the N2 is threat - specific. However, demystification of this finding may come by considering the distinct components of CBT and ABM that may foster different aspects of cognitive control. Specifically, CBT fosters increased activity in frontal - cortical regions during reappraisal (Goldin et al., 2013;2014), a strategy that is meant to be implemented in arousing contexts for anxious individua ls. On the other hand, ABM does not explicitly create a context for learning specific coping strategies in the presence of threat rather , it aims to implicitly train a new learned association . In other words, the aim of ABM is to manipulate attention to focus on neutral faces in the presence of a threatening face (thereby reducing the threatening salience of the emotional face). Speculatively, it could be the case that those in the ABM group learned to engage cognitive control to foster avoidance of emoti onal stimuli to do just that , and in the cases in which this happened, the learned mechanism transferred beyond the treatment program. However, more research is needed to examine if this is the case. Although this study provides insight on change in neurophysiological indices of cognitive processes and their relation s to symptom change , the findings should be interpreted considering a few limitations. Primarily, this study included a small sample, and as such, it would be of great utility fo r future studies to implement similar designs in larger samples to examine if the results remain consistent and achieve greater power for non - significant or marginal effects. In addition, the study did not include a control group assessed throughout the c ourse of treatment . Although the assessment of both treatment groups allo ws for the comparison of change across the interventions, it would also be of value to examine if changes are specific to the interventions themselves, over and beyond the effect of r epeatedly completing the task and the passage of time . Lastly, the dot probe task utilized in this study did not have a neutral - neutral comparison. 35 Future studies should include such a comparison to specify cha n ges in these processes for threat - neutral fa ce conflict in comparison to neutral faces. In sum, the results suggest that both treatment groups produced increases in cognitive control during the dot probe task to both the presentation of faces and the probe. Therefore, it can be concluded that both treatment groups fostered more engagement of frontal brain processes throughout the task related to a global cognitive control mechanism. There were no changes in early attentional processing, which could signify that the malleability of these process es and their measurement is difficult to achieve . Additionally, there was no relationship between the rates of change of ERPs and symptoms throughout the course of treatment. However, there was a relationship between the rate of change among ERPs and among symptoms. Lastly, the results showed that changes in cognitive control at one time - point predicted different effects on symptoms depending on the treatment context . Specifically, a higher N2 predicted more avoidance symptoms in the ABM group, but marginal ly predicted less anxious and avoidance symptoms in the CBT group. More research is needed to delineate a more conclusive interpretation of these findings, however , it is speculated that these differential effects are due to the specific aspects of cogniti ve control that are fostered within each treatment and thereby implemented in the dot probe task. 36 APPENDICES 37 APPENDIX A: T ables 38 Table 1: Descriptive Statistics on the full sample CBT ( n = 33) ABM ( n = 17 ) Age 28.67 (7.1) 27.76 (9.1) % Female 39.4 52.9 % Right - handed 84.8 100 Note. All p Table 2 : Final model for N2 change to face presentation in the Dot Probe Task (Pre - Treatment 8 week mark) Fixed Effect Estimate Standard Error t p Intercept - 2.10 0.44 - 4.75 .000 * Time - 0.41 0.16 - 2.6 .013 * Group - 0.22 0.44 - 0.51 . 61 Group X Time - 0.10 0.16 - 0.63 .53 Variance Components Variance Standard Deviation Covariance -- Intercept 6.06 2.46 Slope (Time) 0. 139 0.37 0.67 Residual 3.77 1.94 Note. * p < .05, ** p < .001 39 Table 3: Final model for N2 change to face presentation in the Dot Probe Task (Pre - Treatment Post Treatment) Fixed Effect Estimate Standard Error t p Intercept - 2.14 0.47 - 4.5 .000 ** Time - 0.38 0.17 - 2.3 .022* Group - 0.22 0.47 - 0.48 .64 Group X Time - 0.11 0.17 - 0.67 .50 Follow - up 0.70 0.87 0.80 .42 Follow - up X Group 0.32 0.87 0.37 .72 Variance Components Variance Standard Deviation Covariance -- Intercept 5.95 2.44 Slope (Time) 0.08 0 .28 0.43 Residual 6.18 2.49 Note. * p < .05, ** p < .001 Table 4: Simple Model for P1 Change Over Time to Face Presentation for Slope Extraction Fixed Effect Estimate Standard Error t p Intercept 2. 71 0. 28 9.597 .000 *** Time - 0. 15 0. 09 - 1.68 . 10 Variance Components Variance Standard Deviation Covariance -- Intercept 2.84 1.69 Slope (Time) 0.18 .42 0 Residual 1.95 1.40 Note. * p < .05, ** p < .01, *** p < .001 40 Table 5: Simple Model for N2 Chan ge Over T ime to Face Presentation for Slope Extraction Fixed Effect Estimate Standard Error t p Intercept - 2.15 0. 43 - 4.99 .000 ** Time - .48 0. 13 - 3.77 . 001** Variance Components Variance Standard Deviation Covariance -- Intercept 5.07 2.38 Slope (Time) 0.20 .44 0 .27 Residual 6.36 2.52 Note. * p < .05, ** p < .01, *** p < .001 Table 6: Simple Model for N1 Change Over Time to Probe Presentation for Slope Extraction Fixed Effect Estimate Standard Error t p Intercept - 1.35 0. 23 - 5.98 .000 ** Time - .02 0. 08 - .33 . 74 Variance Components Variance Standard Deviation Covariance -- Intercept .65 .81 Slope (Time) 0.007 .08 1 Residual 3.03 1.74 Note. * p < .05, ** p < .01, *** p < .001 41 Table 7: Simple Model for N2 Change Over Time to Probe Presentation for Slope Extraction Fixed Effect Estimate Standard Error t p Intercept 3.00 0. 36 8.37 .000 ** Time - .38 0. 16 - 2.36 . 02* Variance Components Variance Standard Deviation Covariance -- Intercept 1.55 1.25 Slope (Time) 0.43 .66 1 Residual 8.41 2.90 Note. * p < .05, ** p < .01, *** p < .001 Table 8: Simple Model for LSAS Total Scores Change Time for the Slope Extraction Fixed Effect Estimate Standard Error t p Intercept 87.70 2.30 38.19 .000 *** Time - 6.98 1.04 - 6.70 . 000*** Variance Components Variance Standard Deviation Covariance -- Intercept 212.38 14.57 Slope (Time) 39.14 6.26 - .14 Residual 83.67 9.15 Note. * p < .05, ** p < .01, *** p < .001 42 Table 9: Simple Model for LSAS Anxiety Scores Change Time for the Slope Extraction Fixed Effect Estimate Standard Error t p Intercept 43.45 1.09 39.96 .000 *** Time - 3.20 .49 - 6.51 . 000*** Variance Components Variance Standard Deviation Covariance -- Intercept 44.76 6.69 Slope (Time) 8.15 2.85 - .12 Residual 23.65 4.86 Note. * p < .05, ** p < .01, *** p < .001 Table 10: Simple Model for LSAS Avoidance Scores Change Time for the Slope Extraction Fixed Effect Estimate Standard Error t p Intercept 44.24 1.41 31.32 .000 *** Time - 3.78 .55 - 6.90 . 000*** Variance Components Variance Standard Deviation Covariance -- Intercept 83.69 9.15 Slope (Time) 10.48 3.24 - .23 Residual 26.22 5.12 Note. * p < .05, ** p < .01, *** p < .001 43 Table 11: Simple Model for SPIN Scores Change Time for the Slope Extraction Fixed Effect Estimate Standard Error t p Intercept 48.16 1.27 38.03 .000 *** Time - 5.36 .61 - 8.81 . 000*** Variance Components Variance Standard Deviation Covariance -- Intercept 70.39 8.39 Slope (Time) 16.05 4.01 - .35 Residual 25.31 5.03 Note. * p < .05, ** p < .01, *** p < .001 Table 12: Bivariate Correlations between ERP and Symptom Slopes 1 2 3 4 5 6 7 8 1. P1 Face Slope -- 2. N2 Face Slope .32* -- 3. N1 Probe Slope .30* .26 -- 4. N2 Probe Slope .23 0 .12 -- 5. LSAS Total Slope - .01 .05 - .09 .09 -- 6. LSAS Anxiety Slope .06 .12 - .02 .12 .97** -- 7. LSAS Avoidance Slope - .07 - .01 - .12 .06 .97** .89** -- 8. SPIN Slope .18 .20 .06 .12 .76** .81** .67** -- Note. p <.1, * p < .05, ** p < .001 44 APPENDIX B: F igures 45 Figure 1 : Flow Chart of Study Design 46 Figure 2 : Topographic representation of the P1 time - locked to the presentation of the faces including all observations for the final sample in the 80 - 150 ms time window. 47 Figure 3 : The effect of Site across time for the N2 to face presentation in the dot probe task. The N2 was largest at site Fz. Figure 4 : The N2 to the faces over time collapsed across both groups. Increased N2 was observed across treatment. 48 Figure 5 : Line Graph of N2 to the faces. The topographic representation depicts the N2 change difference (Pre - Treatment - Post - Treatment). Figure 6 : The relationship between the slopes (pre - post treatment) for the P1 and the N2 to face presentation. 49 Figure 7 : The relationship between the slopes (pre - post treatment) for the P1 to face presentation and the N1 to probe presentation. Figure 8 : The relationship the N2 to face presentat ion and LSAS Avoidance scores, showing that a smaller N2 predicted less symptoms for the ABM group while there was no relationship in the CBT group. * 50 Figure 9: The relationship the N2 to face presentation and SPIN scores, showing that a smaller N2 predicted higher symptoms for the CBT group while there was no relationship in the ABM group. * 51 REFERENCES 52 REFERENCES Amir, N., & Bomyea, J. (2011). Working Memory Capacity in Generalized Social Phobia. Journal of Abnormal Psychology, 120 (2), 504 - 509. doi:10.1037/a0022849 Amir, N., Elias, J., Klumpp, H., & Przeworski, A. (2003). Attentional bias to threat in social phobia: facilita ted processing of threat or difficulty disengaging attention from threat? Behaviour Research and Therapy, 41 (11), 1325 - 1335. Amir, N., Mcnally, R. J., Riemann, B. C., Burns, J., Lorenz, M., & Mullen, J. T. (1996). Suppression of the emotional Stroop effect by increased anxiety in patients with social phobia. Behaviour Research and Therapy, 34 (11), 945 - 948. Association, A. P. (2013). Diagnostic and statistical manual of mental disorders (DSM - 5®) : American Psychiatric Pub. Bantin, T., Stevens, S., Gerlach, A. L., & Hermann, C. (2016). What does the facial dot - probe task tell us about attentional processes in social anxiety? A systematic review. Journal of behavior therapy and experimental psyc hiatry, 50, 40 - 51. Bar - Haim, Y., Lamy, D., Pergamin, L., Bakermans - Kranenburg, M. J., & van Ijzendoor, M. H. (2007). Threat - related attentional bias in anxious and nonanxious individuals: a meta - analytic study. Psychological Bulletin, 133 . doi:10.1037/003 3 - 2909.133.1.1 Batty, M., & Taylor, M. J. (2003). Early processing of the six basic facial emotional expressions. Cognitive Brain Research, 17(3), 613 - 620. Brown, H. M., Eley, T. C., Broeren, S., MacLeod, C., Rinck, M. H. J. A., Hadwin, J. A., & Lester, K. J. (2014). Psychometric properties of reaction time based experimental paradigms measuring anxiety - related information - processing biases in children. Journal of Anxiety Disorders , 28 (1), 97 - 107. Buckner, J. D., Maner, J. K., & Schmidt, N. B. (2010). Difficulty disengaging attention from social threat in social anxiety. Cognitive therapy and research, 34(1), 99 - 105. Bunnell, B. E., Beidel, D. C., & Mesa, F. (2013). A randomized trial of attention training for generalized social phobia: Does attention training change social behavior? Behavior Therapy, 44 (4), 662 - 673. Carpenter, J. K., Curtiss, J., & Hofmann, S. G. The Nature and Treatment of Social Anxiety. Treatments for Psychological Problems and Syndromes, 56 - 73. Clark, D. M. (2005). A cognitive perspective on social phobia. The essential handbook of social anxiety for clinicians , 193 - 218. 53 Connor, K. M., Davidson, J. R., Churchill, L. E., She rwood, A., Weisler, R. H., & FOA, E. (2000). Psychometric properties of the social phobia inventory (SPIN). The British Journal of Psychiatry, 176 (4), 379 - 386. Eldar, S., & Bar - Haim, Y. (2010). Neural plasticity in response to attention training in anxie ty. Psychological Medicine, 40 (4), 667 - 677. Eysenck, M. W., Derakshan, N., Santos, R., & Calvo, M. G. (2007). Anxiety and cognitive performance: attentional control theory. Emotion, 7 (2), 336. Folstein, J. R., & Van Petten, C. (2008). Influence of cogn itive control and mismatch on the N2 component of the ERP: a review. Psychophysiology, 45(1), 152 - 170. Fox, E., Russo, R., Bowles, R., & Dutton, K. (2001). Do threatening stimuli draw or hold visual attention in subclinical anxiety?. Journal of Experimental Psychology: General, 130(4), 681. Goldin, P. R., Ziv, M., Jazaieri, H., Weeks, J., Heimberg, R. G., & Gross, J. J. (2014). Impact of cognitive - behavioral therapy for social anxiety disorder on the neural bases of emotional reactivity to and r egulation of social evaluation. Behaviour research and therapy , 62 , 97 - 106. Grossheinrich, N., Reinl, M., Pogarell, O., Karch, S., Mulert, C., Brueckl, M., . . . Hornig, A. (2013). Effects of Low Frequency Prefrontal Repetitive Transcranial Magnetic Stimu lation on the N2 Amplitude in a GoNogo Task. PLoS One, 8 (6), e67136. Heeren, A., Lievens, L., & Philippot, P. (2011). How does attention training work in social phobia: disengagement from threat or re - engagement to non - threat? Journal of Anxiety Disorders, 25 (8), 1108 - 1115. Heimberg, R. G. (2002). Cognitive - behavioral therapy for social anxiety disorder: current status and future directions. Biological psychiatry , 51 (1), 101 - 108. Heinrichs, N., & Hofmann, S. G. (2005). Cognitive Assessment of Social Anxiety: a Comparison Cognitive Behaviour Therapy , 34 (1), 3 - 15. Huppert, J. D., Roth, D. A., & Foa, E. B. (2003). Cognitive - behavioral treatment of social phobia: new advances. Current psychiatry reports, 5 (4), 289 - 296. Huppert, J. D., Roth Ledley, D., & Foa, E. B. (2006). The use of homework in behavior therapy for anxiety disorders. Journal of Psychotherapy Integration, 16 (2), 128. Julian, K., Beard, C., Schmidt, N. B., Powers, M. B., & Smits, J. A. (2012). Attention training to reduce attention bias and social stressor reactivity: an attempt to replicate and extend previous findings. Behaviour Research and Therapy, 50 (5), 350 - 358. 54 Kessler, R. C., Petukhova, M., Sampson, N. A., Zaslavsky, A. M., & Wi ttchen, H. U. (2012). disorders in the United States. International Journal of Methods in Psychiatric Research, 21 (3), 169 - 184. Klumpp, H., & Amir, N. (2009). Examination of vigilance and disengagement of threat in social anxiety with a probe detection task. Anxiety, Stress & Coping, 22(3), 283 - 296. Klumpp, H., Fitzgerald, D. A., & Phan, K. L. (2013). Neural predictors and mechanisms of cognitive behavioral therapy on thre at processing in social anxiety disorder. Progress in Neuro - Psychopharmacology and Biological Psychiatry , 45 , 83 - 91. Kolassa, I. T., & Miltner, W. H. (2006). Psychophysiological correlates of face processing in social phobia. Brain research, 1118(1), 130 - 141. Krug, M. K., & Carter, C. S. (2010). Adding fear to conflict: a general purpose cognitive control network is modulated by trait anxiety. Cognitive, Affective, & Behavioral Neuroscience, 10(3), 357 - 371. Lavric, A., Pizzagalli, D. A., & Forstmeier, S. ERP components and cortical tomography. European Journal of Neuroscience, 20 (9), 2483 - 2488. Leary, M. R. (1983). A brief version of the Fear of Negative Evaluation Scale. Personality and Social Psychology Bulletin , 9 (3), 371 - 375. Liebowitz, M. R. (1987). Social phobia. Modern Problems in Pharacophychiatry, 22 , 141 - 173. Lundh, L. - G., & Öst, L. - G. (2001). Attentional bias, self - consciousness and perfectionism in social phobia before and after cognitive - be haviour therapy. Scandinavian Journal of Behaviour Therapy, 30 (1), 4 - 16. MacLeod, C., & Mathews, A. (1988). Anxiety and the allocation of attention to threat. The Quarterly Journal of Experimental Psychology, 40 (4), 653 - 670. MacLeod, C., Mathews, A., & Tata, P. (1986). Attentional bias in emotional disorders. Journal of Abnormal Psychology, 95 , 15 - 20. doi:10.1037/0021 - 843X.95.1.15 MacLeod, C., Rutherford, E., Campbell, L., Ebsworthy, G., & Holker, L. (2002). Selective attention and emotional vulnerabil ity: assessing the causal basis of their association through the experimental manipulation of attentional bias. Journal of abnormal psychology , 111 (1), 107. MacLeod, C., Grafton, B., & Notebaert, L. (2019). Anxiety - Linked Attentional Bias: Is It Reliable?. Annual review of clinical psychology , 15 . 55 Månsson, K. N., Carlbring, P., Frick, A., Engman, J., Olsson, C. J., Bodlund, O., ... & Andersson, G. (2013). Altered neural correlates of affective processing after internet - delivered cognitive behavior therapy for social anxiety disorder. Psychiatry Research: Neuroimaging , 214 (3), 229 - 237. Mattia, J. I., Heimberg, R. G., & Hope, D. A. (1993). The revised Stroop color - naming task in social phobics. Behaviour Research and Therapy, 31 (3), 305 - 313. Mogg, K., & Bradley, B. P. (2016). Anxiety and attention to threat: Cognitive mechanisms and treatment with attention bias modification. Behaviour research and therapy , 87 , 76 - 108. Mogg, K., Philippot, P., & Bradley, B. P. (2004). Selective attention to angry faces in clinical social phobia. Journal of Abnormal Psychology, 113 (1), 160. Mogg, K., Waters, A. M., & Bradley, B. P. (2017). Attention bias modification (abm): review of effects of multisession abm training on anxiety and threat - related attention in h igh - anxious individuals. Clinical Psychological Science , 5 (4), 698 - 717. Monk, C. S., Nelson, E. E., McClure, E. B., Mogg, K., Bradley, B. P., Leibenluft, E., ... & Pine, D. S. (2006). Ventrolateral prefrontal cortex activation and attentional bias in resp onse to angry faces in adolescents with generalized anxiety disorder. American Journal of Psychiatry , 163 (6), 1091 - 1097. Moran, T. P., Bernat, E. M., Aviyente, S., Schroder, H. S., & Moser, J. S. (2015). Sending mixed signals: worry is associated with enh anced initial error processing but reduced call for subsequent cognitive control. Social cognitive and affective neuroscience, 10(11), 1548 - 1556. Moriya, J., & Tanno, Y. (2008). Relationships between negative emotionality and attentional control in effortful control. Personality and Individual Differences, 44 (6), 1348 - 1355. Moriya, J., & Tanno, Y. (2011). The time course of attentional disengagement from angry faces in social anxiety. Journal of Behavior Therapy and Experimental Psychiatry, 42(1), 122 - 128. Morrison, A. S., & Heimberg, R. G. (2013). Attentional control mediates the effect of social anxiety on positive affect. Journal of Anxiety Disorders, 27 (1) , 56 - 67. Moser, J. S., Huppert, J. D., Duval, E., & Simons, R. F. (2008). Face processing biases in social anxiety: an electrophysiological study. Biological psychology , 78 (1), 93 - 103. Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortex function. Annual review of neuroscience, 24(1), 167 - 202. 56 Mueller, E., Hofmann, S., Santesso, D., Meuret, A., Bitran, S., & Pizzagalli, D. A. (2009). Electrophysiological evidence of attentional biases in social anxiety disorder. Psycholo gical Medicine, 39 (7), 1141 - 1152. Brain and Cognition , 78 (1), 63 - 73. Price, R.B., Siegle, G.J., Silk, J.S., Ladouceur, C.D., McFarland, A., Dahl, R.E., Ryan, N.D. , 2014. Looking under the hood of the dot - probe task: an fmri study in anxious youth. Depress. Anxiety 31 (3), 178 187. http://dx.doi.org/10.1002/da.22255 Rothbart, M. K., Ahadi, S. A., & Evans, D. E. (2000). Temperament and personality: origins and outcomes. Journal of personality and social psychology , 78 (1), 122. Rotshtein, P., Richardson, M. P., Winston, J. S., Kiebel, S. J., Vuilleumier, P., Eimer, M., ... & Dolan, R. J. (2010). Amygdala damage af specific time windows. Human brain mapping , 31 (7), 1089 - 1105. Santesso, D. L., Meuret, A. E., Hofmann, S. G., Mueller, E. M., Ratner, K. G., Roesch, E. B., & Pizzagalli, D. A. (2008). Electrophysiologica l correlates of spatial orienting towards angry faces: a source localization study. Neuropsychologia , 46 (5), 1338 - 1348. Schmidt, N. B., Richey, J. A., Buckner, J. D., & Timpano, K. R. (2009). Attention training for generalized social anxiety disorder. Jou rnal of Abnormal Psychology, 118 (1), 5. Schmukle, S. C. (2005). Unreliability of the dot probe task. European Journal of Personality , 19 (7), 595 - 605. Schultz, L. T., & Heimberg, R. G. (2008). Attentional focus in social anxiety disorder: potential for i nteractive processes. Clinical Psychology Review, 28 (7), 1206 - 1221. Staugaard, S. R. (2010). Threatening faces and social anxiety: a literature review. Clinical Psychology Review, 30 (6), 669 - 690. Stangier, U. (2016). New developments in cognitive - behavioral therapy for social anxiety disorder. Current psychiatry reports , 18 (3), 25. Telzer, E.H., Mogg, K., Bradley, B.P., Mai, X., Ernst, M., Pine, D.S., Monk, C.S., 2008. Relationship between trait anxiety, prefrontal cortex, and attention bias to angry faces in children and adolescents. Biol. Psychol. 79 (2), 216 222. http://dx.doi.org/10.1016/j. biopsycho.2008.05.00 Thai, N., Taber - Thomas, B. C., & Pérez - Edgar, K. E. (201 6). Neural correlates of attention biases, behavioral inhibition, and social anxiety in children: An ERP study. Developmental cognitive neuroscience , 19 , 200 - 210. 57 Tottenham, N., Borscheid, A., Ellertsen, K., Marcus, D., & Nelson, C. (2002). The NimStim face set. Retreived from http://www . macbrain. org/faces/index. htm . tematic review. Psychophysiology , 55 (6), e13051. Wieser, M. J., Pauli, P., Alpers, G. W., & Mühlberger, A. (2009). Is eye to eye contact really threatening and avoided in social anxiety? An eye - tracking and psychophysiology study. Journal of Anxiety Disor ders, 23 (1), 93 - 103. Woody, S. R., Chambless, D. L., & Glass, C. R. (1997). Self - focused attention in the treatment of social phobia. Behaviour Research and Therapy, 35 (2), 117 - 129. Yiend, J., & Mathews, A. (2001). Anxiety and attention to threatening pictures. The Quarterly Journal of Experimental Psychology: Section A, 54(3), 665 - 681. Zhang, D., Liu, Y., Wang, L., Ai, H., & Luo, Y. (2017). Mechanisms for attentional modulation by threatening emotions of fear, anger, and disgust. Cognitive, Affective, & Behavioral Neuroscience , 17 (1), 198 - 210. Zvielli, A., Bernstein, A., & Koster, E. H. (2014). Dynamics of attentional bias to threat in anxious adults: bias towards and/or away? PLoS One, 9 (8), e104025.