WORRY AND WORKING MEMORY: A BEHAVIORAL & ERP INVESTIGATION ACROSS THE MENSTRUAL CYCLE By Lili anne Marie Gloe A THESIS Submitted to Michigan State University i n partial fulfillment of the requirements for the degree of Psychology Master of Arts 2019 ABSTRACT WORRY AND WORKING MEMORY: A BEHAVIORAL & ERP INVESTIGATION ACROSS THE MENSTRUAL CYCLE By Lilianne Marie Gloe The current project examined the relationship between worry, a component of anxiety characterized by negative, future - oriented thought activity , and working memory in women . Further elucidation of this relationship occurred through us e of m ultiple working memory measures; examining the P300, a n event - related potential measured using electroenceph a l o gram (EEG) thought to index resources available for cognitive processing ; and considering the role of ovarian hormones. It was hypothesized that worry would be associated with poorer working memory function and reduced amplitude of the P300 at higher levels of task difficulty and estradiol . Participants were 65 naturally - cycling women who attended four visits across their menstrual cycle s . On each visit day , data collection included a measure of daily worry (Penn State Worry Que stionnaire), a saliv a sample used to assay for estradiol, and completion of three working memory tasks ( N - back task with EEG recording , Operation Sp an task and Reading Span task) . Five m ultilevel model s were constructed to examine the impact of within - subject fluctuation of Penn State Worry Questionnaire scores and estradiol values on N - back task accuracy, N - back task reaction time, N - back task P300 amplitude, Operation Span score, and Reading Span score. Task parameters of the N - back task (i.e. load and trial type) were included in the three models of the N - back task as indicators of task difficulty . Results indicated that within - subject fluctuations were not significant ly related to working memory performance or the P300 amplitude. Potential reasons for null findings and future directions are explored. iii ACKNOWLEDGEMENTS I would like to thank Deborah Kashy for her valuable consultation with regards to the multilevel modeling structure utilized for the N - back task, as well as the Clinical Psychophysiology Lab undergraduate research assistants, lab managers and graduate stud ents for their contributions to data collection for this project. Additionally, I would like to thank my committee members, Dr. Kat harine Thakkar, Dr. Susan Ravizza and Dr. Jason Moser, for their feedback on this project. iv TABLE OF CONTENTS LIST OF TABLES ... .. v i LIST OF FIGURES .. vii i A Primer on Working Dynamic s pan t asks .2 Complex s pan t asks .3 A Brief Review of the Relationship Between WM and Worry ..3 The Current Study 5 Multiple behavioral measures of WM 5 Neurophysiology 6 Sex an 7 9 METHODS ... 10 . 10 ... 1 1 . 1 1 1 2 1 2 1 2 N - Back t 1 3 O - Span and R - Span t 1 4 Neurophysiological d ata r ecording and p reprocessing .. 1 4 1 7 N - back p erformance and P300 a 18 R - Span and O - Span s core a 0 .. .. .... 21 .... 2 1 .2 1 N - Back . .. 2 1 2 1 2 2 P300 at Pz 300 - ... 23 Span t ask s . 24 R - Span task . 24 O - ... . 24 2 5 30 v A PPENDIX B .. . 43 REFERENCES ......... ... 49 vi LIST OF TABLES Ta ble 1: . .. 37 Table 2: PSWQ score, Estradiol and Span Task Measure Means and Standard Deviations ..... 38 Table 3 : N - back Measure Means and Standard Deviations by Load x Trial Type Interaction 38 Table 4 : Test of Type 3 Fixed Effect Significance for Multilevel Model of Accuracy (%) Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number 39 Table 5: Test of Type 3 Fixed Effect Significance for Multilevel Model of Reaction Time Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number 4 0 Table 6 : Test of Type 3 Fixed Effect Significance for Multileve l Model of P3 at Pz from 300 to 500ms Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number 41 Table 7: T - Scores for Differences between Least Squares Means for P3 at Pz from 300 to 500ms for Each Load by Trial Type Level 42 Table 8: Test of Type 3 Fixed Effect Significan ce for Multilevel Model of R - Span Score related to Estradiol, and PSWQ Score Controlling for EEG Visit Number 42 Table 9: Test of Type 3 Fixed Effect Significance for Multilevel Model of O - Span Score related to Estradiol and PSWQ Score Controlling for EEG Visit Number 42 Table B 1: Residual Correlation & Variance Estimates for Multilevel Model of Accuracy (%) Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number 3 Table B 2: Estimates for th e Intercept and Continuous Variable Coefficients from Multilevel Model of Accuracy (%) Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number . ....4 3 Table B 3: Least Squares Means for Accuracy f or Each Load by Trial Type Level ... ..4 4 Table B 4: Residual Correlation & Variance Estimates for Multilevel Model of Reaction Time (ms) Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number 4 Table B 5: Estimates for the Intercept and Continuous Variable Coefficients from Multilevel Model of Reaction Time (ms) Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number 45 Table B 6: L east Squares Means for Reaction Time for Each Load by Trial Type Level 45 vii Table B 7 : Residual Correlation & Variance Estimates for Multilevel Model of of P3 at Pz from 300 to 500ms Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number 6 Table B 8 : Estimates for the Intercept and Continuous Variable Coefficients from Multilevel Model of P3 at Pz from 300 to 500ms Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number 6 Table B 9: Least Squares Means for P3 at Pz from 300 to 500ms for Each Load by Trial Type Level 7 Table B 10: Estimates for the Intercept and Continuous Variable Coefficients from Multilevel Model of R - Span Score related to Estradiol and PS WQ Score Controlling for EEG Visit Number 7 Table B 11: Estimates for the Intercept and Continuous Variable Coefficients from Multilevel Model of R - Span Score related to Estradiol and PSWQ Score Controlling for EEG Visit Number viii LIST OF FIGURES Figure 1. Ill ustration of study data collection across the menstrual cycle . Day 0 is menstruation. For the purposes of this study, only estradiol levels will be considered. The N - back, O - Span, and R - Span tasks are completed 31 Figure 2. Subject and session loss for each dependent measure of the study 32 Figure 3. Stimulus - locked grand average waveforms at Pz for a) non - targets, b) targets, and c) lures. St imulus presentation occurred at 0ms 33 Figure 4. Breaking down the 3 - way PSWQ Score x Estradiol x Load interaction identified in the multilevel model for accuracy (a) At high levels of estradiol (+1SD), there is no significant interactions between load and PSWQ Score. (b) At low levels of estradiol ( - 1SD), there is a PSWQ Score is significantly related to accuracy for 3 - back blocks, such that increased worry predicts worse accuracy. 1 INTRODUCTION Sex disparities in anxiety have long been recognized. W omen suffer from higher rates and longer course of anxiety than men (Kessler, Petukhova, Sampson, Zaslavsky, & Wittchen, 2012; McLean, Asnaani, Litz, & Hofmann, 2011) . Despite the disproportionate number of women suffe ring from the effects of anxiety , studies have largely failed to examine the precise nature and im pact of anxiety in women . Thus, t he current study focuse d on the relation between anxiety and one important aspect of functioning previously linked to anxiety cognition in women. M odels of anxiety and cognition are currently stated in general terms and are intended to apply to all people, regardless of sex. One prominent model of the relationship between anxiety and cognitive impairment is Attentional Control Theory (ACT; Eysenck & Calvo, 1992; Eysenck & Derakshan, 2011; Eysenck et al., 2007) . The theory asserts that anxie ty, via worrisome thoughts that load the cognitive system, depletes available working memory resources. To overcome, or compensate for, this dearth of working memory resources, ACT proposes that anxious individuals recruit auxiliary resources so that they can complete the task at hand (Eysenck & Calvo, 1992; Eysenck et al., 2007) . As a result, they de monstrate inefficient performance, expending greater resources to complete the task at a comparable level of accuracy as non - anxious individuals ( Eysenck et al., 2007) . While ACT has garnered empirical support (e.g. Eysenck et al., 2007) , the literature is limited and heterogeneous. In particular, whether and how worry a component of anxiety comprised of negative, future - oriented verbal thought activity (Borkovec, Ray, & Stober, 1998) - - is related to deficits in working memory is unclear. T he applicability of this theory to women has also been assumed, but has not been explicitly examined . In addition to being central to 2 ACT, understanding how worry relates to working memory is particu larly critical for women , given that women have been shown to experience higher levels of worry than men (Nitschke, Heller, Imig, McDonald, & Miller, 2001) . Therefore, the current study aimed to better characterize the relationship between worry and working memory in women. To further elucidate this important rel ationship, this study 1) utiliz ed several working memory tasks , 2) includ ed behavioral and neurophysiological measures , and 3) consider ed the role of ovarian hormones in women across the menstrual cycle . A Primer on Working Memory Working memory (WM) is a multi - component system to simultaneously st ore and manipulate information during tasks (Baddeley, 1996 ; Baddeley, 1986) . WM - in that it allows goal - relevant information to be readily available despite competing distract o rs (Kane, Conway, Hambrick, & Engle, 2007) . Given that WM is a multi - faceted concept, many tasks have been created to measure it . For the purposes of this brief review, I will focus mostly on literature that uses two common types of tasks: dynamic span and complex span tasks. Dynamic span tasks. Dynamic span tasks involve continuous attention to a series of items presented one after another , and the updating of WM to reflect relevant target items during presentation (Moran, 2016). The N - back task is one of the most common ways in which to measure dynamic span. It requires participants to respond when an item is shown that had been presented N trials previously, with N defined for each block of trials. N - back tasks exist in several modalities, including a verbal version in which participants res pond when a particula r letter i s shown visually that matched the one shown N trials earlier, and a spatial form, in which participants 3 respond when the letter is shown in the same location as was shown N trials earlier. In addition to manipulating task load by varying the numb er of trials back the target letter occurred in the sequence (i.e. N ) , variations of this task have also modulated task difficulty through the inclusion of lure trials within the sequences. Lures are letters that match a recently shown, but non - target lett er. Lure trials are thought to produce more interference than other non - targets , as inhibiting responses to lure s requires more cognitive control because of the familiarity/saliency of stimuli (Gray, Chabris, & Braver, 2003) . Complex span tasks. In complex span tasks, participants are presented with a series of letters interleaved with a demanding secondary task (Moran, 2016) . Two popular complex tasks are the Operation - Sp an ( O - Span; Turner & Engle, 1989; Unsworth, Heitz, Schrock, & Engle, 2005) and Reading - Span task s ( R - Span; Daneman & Carpenter, 1980) , in which participants are presented with letters interleaved with mathematical operation problems (O - Span) or sentence comprehension exercises (R - Span) for a sequence of trials . At the conclusion of the trial sequence, they are asked to recall the presented letters in perfect order . The sum of the number of letters contained within perfectly recalled sequences are used to produce a WM capacity score. A Brief Review of the Relationship B etween WM and Worry Although a great deal of work has investigated the relationship b etween anxiety and WM processes , the nature of the more specifi c relationship between worry and WM processes is still un c lear . A recent meta - analysi s examined the worry and WM relationship for tasks in the spatial (k=3, N=258) and phonological domain (k=7, N=647) and found moderate and small effect sizes , respectively (Moran, 2016) . However, in examining the worry and WM literature more closely , the few studies that have examined this relation ship are very diverse in methodology . 4 They utilize a variety of WM tasks, worry measures, sample composition s ( e.g. clinical v. healthy participants, age ), and study design s (e.g. worry induction vs. trait worry). Because variability in the methods and sample composition is high , especially in proportion to the number of studies of worry and WM overall , it is important to understand the evidence for thi s relationship in the context of individual studies . In healthy populations, studies utilizing dynamic span tasks have found relationships between trait worry and lower accuracy on 2 - back and 3 - back version s of the N - back task (Bredemeier & Berenbaum, 2013; Moran, 2016) . Additionally, one study of participants diagnosed with General ized Anxiety Disorder (GAD) a disorder characterized by excessive worry (American Psychiatric Association, 2013) showed that GAD patients had increased reaction times on 2 - back and 3 - back correct target trials of a verbal N - back task (Stefanopoulou, Hirsch, Hayes, Adlam, & Coker, 2014) . Thus, although there appears to b e a relationship between worry and dynamic span performance, the way in which worry impacts behavior al performance is unclear. A more obscure relationship emerges in studies utilizing complex span tasks . In the same study that found a relationship between lower N - back accuracy and trait worry, Bredemeier & Berenbaum ( 2013) found no relationship between worry and O - Span score . In contrast , Trezise & Reeve ( 2014) found that state testing worry was associated with lower accuracy on a modified O - Span task that used arithmetic problems as target stimuli in adolescent students . Ganley & Vasilyeva ( 2014) further showed that testing worries were related to poor performance on two different span tasks . However, in a second study , which included a larger sample , a more balanced sex distribution , and was conducted at a different university, they only observed a relationship between worry and poor visuospatial WM (Ganley & Vasilyeva, 2014) . Although it 5 is unclear why the relationship between worry and complex span task performance differ ed across studies , it could be related to the sample characteristics (i.e. adults v. adolescents; female dominated v. equal sex distribution), the difference in worry measures (i.e. tra it v. state), and/or the use of different tasks and stimuli . The Current Study Despite previous literature demonstrating that the relationship between worry and WM appears to be present across many contexts , the heterogeneity in results indicate s the need for further work to understand the precise nature of this relationship. This need was addressed in the current study by considering multiple measures of WM , a neurophysiological index of WM and the role of ovarian hormones in women across the men strual cycle . Multiple b ehavioral m easures of WM . First, I examine how worry is related to performance on three measures of WM . Specifically , I have include d a verbal N - back task and two complex span tasks in the analysis to more fully capture WM as multi - modal construct in its relationship with worry. In exploring dynamic and complex span tasks together , recent investigations have found that, while both tasks seem to assess general WM , unique task - specific factors provide additional information that cannot be yielded through the use of a single WM task alone. For instance , results from a meta - analysis found weak correlations between N - back and complex span performance (Redick & Lindsey, 2013) . However, in response to the finding s of Redick & Lindsey (2013), a study using latent - variable analysis found strong correlations between task types after controlling for measurement error and task content (e.g. letters vs. numbers; Schmiedek, Lövdén, & Lindenberger, 2014 ) . At the neural level, Minamoto et al. ( 2017) found that the lateral prefrontal cortex (lPFC) appears to be implicated in task - related functional connectivity across both N - back 6 and complex span tasks. H owever, the regions that are recruited in tandem with the lPFC differ by task and N - back content (Minamoto et al., 2017) . Together, th e extant literature suggests that dynamic and complex span tasks index a general aspect of WM while simultaneously assessing elements of WM that are task - specific. Because task - specific variance is important to consider in studies of WM , I have utilize d both dynamic and complex span tasks in the current analysis to allow for assessment of task - general and task - specific effects of worry on WM . Task - general WM is more global and does not depend on the content of the to - be maintained/manipulated items, while task - specific WM is specif ic to the content of such items and can be divided into verbal/p honological and visuospatial content categories. R ecent work has provided evidence for a relationship between worry and reduced task - general WM (Moran, 2016) . However, the verbal nature of worry suggests it could also relate to poorer processing of verbal/phonological items , in particular , above and beyond general WM defi cits (Borkovec et al., 1998; Moran, 2016) . Importantly, the tasks used in the current study all utilize verbal stimuli to be manipulated. It should be noted that R - Span task may introduce additional interference in verbal - related processes because the secondary task involves verbal material, as opp osed to the numerical operations utilized by the O - Span task. Neurophysiology . Second, a lthough exami ning behavioral performance on several WM tasks will contribute to a fuller understanding of how worry may relate to poor WM function , additional information about the nature of this relationship can be gleaned from neurophysiological investigation. To date, no work has directly examined the relationship between worry and neural correlates of WM . 7 Using electroencephalogram (EEG), a direct measure of online electrical brain activity, neural responses during WM tasks can be examined with high temporal precision. These neural responses, called Event - Related Potentials (ERPs), consist of deflections of voltage that occur in response to a variety of external (e.g. presentation of a digit) and internal (e.g. commission of an error) stimuli and reflect specific neural or psychological processes (Kappenman & Luck, 2011) . Although the relationship between worry and WM deficits has not be en examined using neurophysiological methods, there is precedence for recording ERPs during the N - back task. Past work has indicated that the amplitude of the P300 (or P3b) ERP component is modulated by level of load in the N - Back task (Bailey, Mlynarczyk, & West, 2016; McEvoy, Smith, & Gevins, 1998; Watter, Geffen, & Geffen, 2001; West, Bowry, & Krompinger, 2006) . The P 300 is a positive voltage change maximal at parietal sites that peaks approximately 300 800ms after a task - relevant, salient, or novel stimulus is presented. The P300 can be thought of as an index of attentional resource allocation, such that it is increased when more attention is devoted to the relevant extern al stimuli (i.e. target trials; Polich, 2012) . This interpretation can be a pplied to the existing literature demonstrating that the P300 is reduced in amplitude as memory load increases in the N - Back (Bailey et al., 2016; McEvoy et al., 1998; Watter et al., 2001; West et al., 2006) . At higher load s , there is a great reliance on WM processes , because more stimuli must be held in WM and updat ed as new stimuli are presented . Thus, the reduced P300 amplitude at higher loads reflects fewer available resources for processing the external stimuli , because such resources are being devoted to internal WM processes. Because worry puts further strain on WM resources, the P300 should be even smaller at higher loads in worriers . 8 Sex and o varian hormones. Finally, the current study also consider s the relationship between worry and WM in women as a function of hormonal status across the menstrual cycle . It is possible that the heterogeneity of previous worry and WM findings may also be due, in part, to ignoring s ex differences and/or fluctuations of ovarian hormones across the menstrual cycle. It is critical to consider the role of ovarian hormones in studies of cognition and anxiety, given their role in both Maeng & Milad, 2015; Man, MacMillan, Scott, & Young, 1999; Montoya & Bos, 2017) . Animal w ork by Shansky and colleagues have found that female rat WM performance suffers under acute stress at high levels of estradiol . The worst performance is found in (1) naturally cycling rats d uring the proestrus phase where estradiol is high -- compared to rats in phases characterized by lower estrogen levels and (2) in overiectomized rats given a long - term estrogen replacement (Shansky & Lipps, 2013 ; Shansky et al., 2004; Shansky, Bender, & Arnsten, 2009 ; Shansky, Rubinow, Brennan, & Arnsten, 2006) . Hypothesized mechanisms by which high estradiol may negatively impact WM performance under stress include the intensification of the effects of glutocorticoid release , increases in the availability of dopamine , and disruption of the balance between dopaminergic and noradrenergic receptor activation (Shansky & Lipps, 2013) . Some human work has also reported deleterious effects of stress on WM in women (Schoofs, Pabst, Brand, & Wolf, 2013) , but findings have been inconsistent and have not examined the role of ovarian hormone s (Shields, Sazma, & Yonelinas, 2016) . Moreover, despite th e examination of stress and WM in females , to my knowledge, no studies have examined the role of ovarian hormones in the relationship between worry and WM . 9 Hypotheses. The primary hypothesis of the current study was that worry w ould be associated with poorer working memory function at higher levels of difficulty and estradiol , because it is under these conditions ( Eysenck et al., 2007; Shansky & Lipps, 2013) . It was expected that worry w ould be related to deficits in WM performance (either in longer RT or reduced accuracy) and a reduced P300 amplitude when the task wa s most difficult (i.e. at higher loads and more difficult trial ty pes of the N - back ) , but only when estradiol wa s high . Because it was N - back task is task - general, t wo hypotheses for how estradiol may be implicated in the relationship between worry and complex span task performance were considered. If worry relates to task - general WM, then worry would be related to poorer performance on both span task s when estradiol is high . If worry has a particularly strong effect on phonological/ve rbal WM , then worry would be related to poorer performance on the R - Span only . 10 METHODS Participants Participants were 6 7 female volunteers 18 to 25 years old recruited for the MSU Clinical Psychophysiology Lab Brain Cycle Study . Participants were recruited from East Lansing, Lansing and surrounding areas in Michigan via commercial mailing lists, paid and public service advertisements in local media, flyers, and online advertisements via Craigslist. Demographic information about the sample is p rovided in Table 1. Overall, the sample was predominately white, heterosexual and had a gender identit y of female. However, not ably, the sample also co nsisted of several participants who belong to racial and sexual orientation groups that are traditionally under - represented /reported in the literature. Because the Brain Cycle Study aims to examine cognitive impacts of worry in women across the menstrual cycle, many inclusion/exclusion criteria are in place to ensure other exogenous and endogenous factors are not acting upon endocrine system functioning. Thus, to be eligible for the study, participants had to be naturally menstruating (i.e. every 22 - 35 days) and not taking hormonal contraceptives, psychotropic medications and steroid medications during the pas t eight weeks before study participation. Women must have had no history of genetic or medica l conditions known to have an impact upon the endocrine system. Additionally , those who ha d epilepsy; have hearing, visual, or other physical or mental impairments that could interfere with data quality; or ha d experienced head trauma that resulted in a loss of consciousness for over five minutes were also excluded from the study because of potential effects on neurophysiological data collection. Eligibility was confirmed during each visit of the study to ensure criteria was consistently met. 11 Procedure Overview. The Brain Cycle Study consists of daily questionnaires and saliva sample collection, one intake visit, four EEG visits and a final visit for administration of a structured clinic diagnostic interview. An overview of data collection is provided in Figure 1. Volunteers interested in participating in the study were screened over the phone for eligibility using the aforementioned crite ria. Menstrual cycle history w as also collected during the phone screen so that the study staff could schedule participants for EEG visits that correspond ed to phases of their menstrual cycle, enabling the collection of data across the entire menstrual cycle for each participant. The t iming of the first EEG visit was randomly selected based on their current menstrual cycle phase (i.e. early follicular phase, late follicular phase, ovulation phase, or mid - luteal phase) to ensure that a similar number of participants start ed in each of the four phases. Eligible participants came to the Clinical Psychophysiology Lab (CPL) in the MSU Psychology Building f or an intake visit that consisted of confirmation of eligibility criteria, orientation to study procedures, the provision of important information regarding daily data collection procedures (e.g. the website address for daily online questionnaire assessments), a nd carrying out written consent procedures. Participants were compensated $280 for full participation, with prorated compensation provided for partial data collection. Daily q uestionnaire p rocedure. Participants were asked to fill out a series of questionnaire s between 5:00PM and 10:00PM each day via Qualtrics, an online assessment portal. Paper c opies of the questionnaire s were provided to each participant in case they are unable to acces s the Internet during this 12 time frame. The daily questionnair e s consisted of the Penn State Worry Questionnaire (PSWQ; Meyer et al., 1990) and other assessments of anxiety, depressio n and eating pathology th at were n ot included for the current analysis. The PSWQ asked about worry during the day that the questionnaire was completed. Daily s aliva s ample c ollection. After being provided with collection tubes at th e intake visit, participants were taught to use the passive drool method to provide their daily samples. Participants were instructed to provide their samples within thirty minutes of waking and were asked not to eat, drink , brush their teeth, or smoke before samples are taken . Samples we re provided for 35 consecutive days in order to capture estradiol levels across the menstrual cycle. Participants were instructed to store their samples in their home freezer immediatel y after daily collection and were provided with materials to ensure the ir samples did not thaw when tra nsported to the lab. Samples were logged and stored in a lab - 80 ° F freezer until they were shipped to Salimetrics , LLC (State College, PA) . Saliva was analyzed using enzyme immunoassay kits for assaying estradiol as specified by (Klump et al., 2016) . Of samples collected on EEG visit days, a 99% retention rate was achieved for estradiol assay. EEG v isit p rocedures. Each participant report ed to the CPL in the MSU Psychology building for each EEG visit, estimated to last two to three hours in duration. After confirmin g eligibility, study staff complete d set - up for EEG recording. The participant then complete d a battery of cognitive assessments on the computer, which consist ed of a Flanker task (n ot included in the current analysis ), followed by verbal N - back task and automated O - Span and R - Span tasks. The O - Span and R - Span tasks were counterbalanced for order of administration. EEG recoding was only 13 conducted during the Flanker and N - back task. At the en d of the visit, the participant complete d an online questionnaire assessment. This assessment contain ed a series of questionnaires assessing for a wide variety of attitudes, beli efs and symptomology that were not used in the current analysis. N - Back t ask . The verbal N - back task, constructed by 2011) , involves the presentation of a continuous stream of letters one at a time in a time - locked fashion. Each letter is display ed for 1000ms, with an ITI of 1 100ms. Participants were asked to respond within 2000ms of the stimulus appearing using the two mouse buttons, with each mouse button corresponding to targets and non - targets. M emory load was manipulated such that the target for each block of trials is identified as the letter shown n trials previously in the block. Targets were 0 - back, 2 - back or 3 - back letters. During 0 - bac k trial blocks, participants were asked to respond to the target lette In 2 - back or 3 - back trial blocks, the target letter was the letter that appeared two or three trials, respectively, before the current trial. For example , in the sequence A - T - R - T - C, the second T in the sequence is the target in a 2 - back block . The task consist ed of 16 blocks with a total of 320 trials ( 0 - back = 160 trials, 2 - back = 80 trials, 3 - back = 80 trials). In addition to standard target and non - target trials, a small subset of non - target trials presented during 2 - back and 3 - back blocks consist ed of lure trials. L ures are letters that match a recently shown, but non - target , letter. As an example, consider the sequence K - F - E - D - K shown during a 3 - presented four letters back, not three letters back. Participants were counterbalanced to one of four ver s ions of the N - back using a Latin Square design . 14 N - back data was preprocessed in Matlab. Practice trial were removed from the data set. Trials were included in calculations of reaction time and accuracy if the reaction time was greater than 200ms. Reaction time was only calculated for trials on which correct respons es were given. O - Span and R - Span t asks . The automated O - Span task (Turner & Engle, 1989; Unsworth et al., 2005) consisted of a series of mathematic operation problems , whereas the automated R - Span task (Daneman & Carpenter, 1980) involved a series of comprehension exercises. In between each math problem or sentence comprehension, a letter was presented. At the end of the series of trials, partic ipants were required to recall the letters in their presented order. The number of letters presented ranged from 3 to 7 for each series of trials. For each task, a final score was calculated as the total number of letters recalled in series for which recal l was perfect for all letters in the series . The maximum final score was 75. Scores were calculated in E - Prime. Neurophysiological d ata recording and p reprocessing . Continuous electroencephalographic (EEG) activity was recorded during the N - back task using the ActiveTwo Biosemi system (BioSemi, The Netherlands) from 64 Ag - AgCl electrodes fitted in a stretch - lycra cap . The location of the c ap electrode ports is based on the 10 - 20 system . The - 20 refers to the standardized method of placing each of the scalp electrodes each electrode is spaced apart from adjacent electrodes at a distance of either 10% or 20% of the total front - back to right - left distance of the skull . Measurements we re taken to ensure proper cap fit, with cap size determin ed by the distance from the nasion (the distinctly depressed are a between the eyes) and the inion (the lowest point of the skull on the back of the skull identified by a prominent bum p ). Centering of the cap was achieved by measuring the distance between the 15 ears around the top of the head, with the tip of each ear being used as a meas urement endpoint. A chin strap was used to hold the cap in place in a tight, but comf ortable fashion. Electrodes were plugged into each of the labeled ports , with labels consisting of combinations for letters and digits (e.g. Pz, C2, T7). The first letter of the label corresponds to areas of the cerebral cortex (i.e. F = frontal, T= temporal, C= central, P = parietal, and O = occipital lobes) . The second part of the label can either be a letter or number and indicates location on the scalp in relation to midline sites. The letter indicates a location along the midline of the scalp, while odd numbers indicate left hemisphere sites and even numbers indicate right hemisphere sites. Sensors wer e also placed on the left and right outer canthi (the outer corners of the eyes where the upper and lower lids meet) and below the left eye (approximately 1cm from the pupil) . Activity recorded from the FP1 site and the three external eye sensors were used to measure electrooculogram (EOG) activity resulting from blinks and eye - movements. Two sensors were also placed on the left and right mastoids bone protrusions behind the ears to use during offlin e analyses as references. The Common Mode Sense (CMS) active electrode and the Driven Right Leg (DRL) passive electrode form ed the ground during data acquisition. In addition to acting a s a reference, the CMS - DRL loop ensures that the average voltage of th e participant stays within a reasonable range, thereby limiting current that could potentially return to the participant. All signals were digitized at 1,024Hz, which represents 1,024 samples of data taken per second and provides millisecond precision . O ffline Analyses was conducted with BrainVision Analyzer 2 (Brain Products, Gilchi ng, Germany). Recordings were vi sually examined by L. Gloe to determine if problems during EEG recording resulted in significant artifactual noise at any recording channel . If a small number (less than or equal t o five) channels were found to contain obvious noise, these channels were 16 removed and interpolated based on activity in the channels closest to the removed channel on the scalp. I f greater than five channels needed to b e interpolated for adequate data to be obtained, the p t included in analyses and was considered missing. Recordings were band - pass filtered with cuto ffs of 0.1 Hz and 30Hz (12dB/oct roll off) and re - referenced to the numeri c mean of the mastoids. O cular correction was then conducted using a common regression method (Gratton, Coles, & Donchin, 1983) . This method accounts for eye movement and blink s , which are typically at their greatest magnitude at frontal sites of the scalp that are near the eyes . Additionally, the method includes calculation of a propagation factor that estimates the differential impact of these movements at sites across th e scalp . The recordings were then segmented based on cognitive load (0 - back, 2 - back, 3 - back) and stimulus type (target, nontarget, lure) on correct trials only error trials were removed. Segments were made relative to stimulus presentation, such that seg ments begin 200ms prior to each letter stimulus and terminate 1,000ms post - stimulus onset. For each set of segmen ts, artifact correction was carried out using a n algorithm such that trials were rejected if they contain ed the any of the following: activity characterized by a voltage step greater than 50 microvolts/ms compared to both the preceding and following trials, respectively; a 200ms time window with a difference in voltage of 300 microvolts; a 100ms time window in which a difference in voltage was le ss than 0.5 microvolts; or an amplitude more extreme than +/ - 200 microvolts. Within in each segment, acti vity in individual channels was averaged across trials, resulting in a single average for each channel of each segment for each participant. Baseline c orrection subtract ed the average activity 200ms prior to stimulus onset from each data point after stimulus onset for each trial type . These averages for each participant were - aver 17 The P300 was then scored in the 300 - 500ms time window at Pz, where it reached maximal amplitude . Analysis Procedure Data were cleaned and prepared for modeling in R (Version 3.4.2) . PSWQ score and estradiol were person - mean centered, such that the mean for each measure was calculated across visit s for a subject a score on each day. Thus, significant effects involving these measures in models can then be interpreted in terms of fluctuations within a person instead of in te rms of differences between people . Additionally, accuracy was examined across each load - by - trial - type interaction level and was labeled as missing for a level if it fell below 30%, as this is believed to be an indicator of lack of effort or misunderstandin g task directions. Multilevel models were utilized in order to examine the impact of worry and estradiol on N - back task accuracy, N - back task reaction time, the P300 elicited during the N - back, O - Span Score, and R - Span score over four EEG visits across the menstrual cycle . All modeling was executed in SAS software (Version 9.4; SAS System for Windows) to accommodate the necessary model structure. For all multilevel models, a random intercept was included to account for nonindependence due to repeated measur es of behavioral measures, the PSWQ, estradiol, and P300 amplitude from each subject. Intra - class correlations (ICCs) were calculated for each model to provide a measure of the extent to which behavioral measures and the P300 amplitude from the same subjec t were correlated. Additionally, the proportion of variance explained by each model ( i.e. marginal R 2 ) was calculated using residual variance from the base/unconditional model, which contained no fixed effects and contain ed only the intercept , and comparing it to that of the full mo del for each dependent variable. The following formula was utilized: R 2 = 18 , where represents the residual variance of the base model and represents the residual variance of the full model. For the fixed effects of all multilevel models constructed, the effect of EEG visit number was included as a covariate in order to account for any practice effects that may have occurred as a result of completing the N - back and span tasks multiple times . The significance of each variable and their interactions was assessed using a T ype III test for fixed effects, which test degree of unique variance explained by a categorical predictor over and above al l other variables in the model. N - back performance and P300 a nalyses . For the N - back task, three models were constructed that varied by their dependent variable, such that there was a model for accuracy, reaction time, and the P300 at Pz from 300 500 ms . The site and time window for the P300 were selected based on common sites examined in the literature and based on visual examination of grand - mean averages of the data. The random effects for the models of the N - back task were constructed to account for heterogeneity in variance across and covariances between measures at each of the eight levels of the load - by - trial - type interaction (i.e. 0 - back non - target, 0 - back target, 2 - back non - target, 2 - back target, 2 - back lure, 3 - back non - target, 3 - back target, 3 - back lure) across each EEG visit within each p articipant . Load - by - tri al type levels occurring within the same visit and subject would be expected to be more similar than those occurring in separate visits or in different subjects. An unstructured covariance matrix was utilized, given evidence of considerable differences in variances and covariances within and between load - by - trial - type interaction levels, respectively . In order to calculate ICC s and the proportions of variance explained by the full model (i.e. R 2 ) , 19 an estimate for residual variance was obtained by averaging the estimates of variance for each of the eight trial types. For each N - back model, a 4 - way interaction was tested between load, trial type, estradiol and PSWQ score s in order to understand the impact of estradiol and PSWQ score s on the N - back task measures at different task difficult. Examining this interaction allowed me to test the hypothesis that higher levels of worry measured via PSWQ scores would be related to increas ed RT, decreased accuracy , and a smaller P300 at higher levels of load (i.e. 2 - back and 3 - back loads) and on more difficult trial types (i.e. target and lure trials; see Gray, Chabris, & Braver, 2003) when estradiol was high. All lower - order terms were inc luded in each model in addition to a main effect of E EG visit number. E stimates for load, trial type and their interaction were derived from the least squares means from the full model, which represents the partial means for each level of each variable/interaction while holding all other variables in the model at their mean . Estimates for estradiol, PSWQ score s , and their interaction were evaluated by running the sam e full model using effects - coded predictors for trial type and load. Any significant interactions between continuous and categorical variables were broken down by simple slopes analyses in which separate intercepts and slopes for the continuous variable we re computed at each level of the categorical variable(s) involved. Any interaction involving both estradiol and PSWQ score was followed - up by using the procedures described by Aiken, West, & Reno ( 1991) . Simple slopes analyses were conducted to examine the effects of one continuous variable at high (+1SD) and low ( - 1SD) values of the other continuous variable. 20 R - Span and O - Span score a nalyses . For the models of span scores, the interaction between estradiol and PSWQ was tested along with its lower order terms and a main effect of EEG visit number. Models were created with person - mean centered PSWQ scores and estradiol as covariate s . If worry impacts task - general WM , PSWQ was expected to be related to reduced O - Span and R - Span scores when estradiol is high . On the other hand, if worry is specifically implicated in verbal/phonological - specific WM , PSWQ score would be related to reduced R - Span scores at higher levels of estradiol . There would be no relationship between worry and O - Span scores in this scenario . 21 R ESULTS Data Retention Figure 2 depicts a flowchart of the data loss across the various measures. T he number of subjects with at least three sessions worth of data for analyses was 61 for N - back accuracy and reaction time, 59 for EEG analyses, and 61 for O - Span and R - Span. For the sake of power, subjects with sessions missing daily estradiol and PSWQ Score on some (but not all) visit days were retained for analys e s, although their data on the missed day was not analyzed. Descriptive Statistics for Predictors and Dependent Variables Descriptive statistics for PSWQ scores, estradiol, and R - Span and O - Span scores are depicted in Table 2. Descriptive statistics for N - back measures are provided are p rovided in Table 3 . ERP w aveforms for each trial type at Pz are presented in Figure 3 . N - Back Task Accuracy . Results of the Type III tests for the fixed effects of the full model predicting accuracy are presented in Table 4 . Firs t, the influence of trial type and load on N - back accuracy was observed . By examining the effect of task - variables on performance irrespective of PSWQ scores and estradiol, a clearer understanding of how the N - back task functi ons over time can be achieved. Notably, variances at each trial type - by - load level differ greatly from one another, as do the correlation of accuracy between levels over time (Table B 1). This suggests that accuracy at some levels of the task changes more across time than at other levels and that accuracy at levels of the task are uniquely related to each over time. Model estimates for main effects and least square means for trial - by - load interaction levels are displayed in Tables B2 and B 3 . Significant differences in accuracy were found between all trial type ( p <0.001) except between 2 - back non - 22 targets and 3 - back non - targets ( t ( 223 ) = 1.99 , p = 1.00 ) , 2 - back target and 2 - back lures ( t ( 220 ) = 2.42 , p = 0.453 ) , 3 - back targets and 3 - back lures ( t ( 212 ) = 0.02, p = 1.00) . Accuracy significantly reduce d as load increase d for targets and lures, while accuracy on non - target trial s did not differ between 2 - back and 3 - back loads. As expected , increased load lead to worse performance on more difficult trial types (i.e. targets and lures), but not on the easier non - target trials. In examining interactions involving PSWQ score and estradiol , marginal interactions PSWQ x estradiol and PSWQ x est radiol x load emerge d . In breaking down the latter three - way interaction, it was discovered that this interaction was driven by a significant simple slope for PSWQ Score s at low levels of estradiol on 3 - back trial s ( B = - 0.2159, SE = 0.096, t (209) = 0.025) . A s depicted in Figure 4 , higher PSWQ scores were related to lower 3 - back accuracy in the presence of low e stradiol levels. Results support the hypothesized relationship between higher PSWQ scores and lower N - back performance on harder trial types, but are in direct contrast to the hypothesis that such a relationship should emerge in the presence of high estradiol levels . Reaction t ime. Type III test of significance for the full model for reacti on time are pr esented in Table 5 . First, the influence of trial type and load on N - back reaction time was observed. Similar to the model for accuracy , residual variances at each trial type - by - load level differ greatly from one another, as do the covariances between residuals of levels over time (Table B 4). This suggests that reaction time at some levels of the task changes more across time than at other levels and that reaction time at levels of the task are uniquely related to each overtime. Model main effects and least square means for tr ial - type - by - load interaction levels are displayed in Tables B5 and B 6. As shown in Table 5, significant effects of trial type, load and their interaction emerged from 23 the full model. Significant differences were found between all trial type s ( p <0.05) except between 3 - back targets and 2 - back lures ( t (225) = - 1.03, p = 1.00), 2 - back target and 3 - back lures ( t (225) = - 1.84, p = 1.00), 3 - back targets and 3 - back lures ( t (225) = 2.16, p = 0.8985). Together, results suggest that participants respond m ore slowly as load increases for non - targets and targets and respond most quickly on non - target trials compared to targets and lures. This pattern of responding confirms my expectations in terms of more slow responses for more difficult trial types , althou gh the lack of significant differences between 2 - back and 3 - back loads for targets and lures was surprising. Importantly , the effect of estradiol and PSWQ score was also examined in the model . Contrary to my hypothesis, however, PSWQ score s and estradiol levels were not significantly related to N - back reaction time. P300 at Pz 300 - 500 ms . Results of Type III test of significance for the full model for the P3 00 in the time window 300 - 500ms at s ite Pz are presented in Table 6 . First, the influence of trial type and load on N - back reaction time was observed. Similar to reaction time and accuracy models , variance and correlation estimates across time were heterogeneous by load by trial type interactions (Table B 7 ) . Again, this suggests that P300 amplitude at some levels of the task changes more across time than at other levels and that P300 amplitude at levels of the task are uniquely related to each over time. Model estimates for main effects and least square means for trial type - by - load interaction levels are displayed in Tables B8 and B 9 . Significant differences between least squares means for the levels of this interaction are presented in Table 7 . As expected, the P300 amplitude was significantly reduced on 2 - bac k and 3 - back loads compared to 0 - back load for targets and on 3 - back load compared to 2 - back load for non - targets . However, in contrast to 24 expectations, P300 amplitude did not significantly differ between 2 - back and 3 - back target s nor between 2 - back and 3 - back loads compared to 0 - back loads for non - targets . Additionally, in contrast to expectations , load did not influence P300 amplitude on lure trials . Importantly , the effect of estradiol and PSWQ score was examined. Contrary to hypotheses , PSWQ score s a nd estradiol levels were not significantly related to P300 amplitude. Span Tasks R - Span t ask . Results of the full model for R - Span score are presented in Table 8 and estimates for the intercept, estradiol, P SWQ score and their interaction are depicted in Table B 10 . No variables of interest emerged as significant predictors of R - Span score . In contrast to both hypotheses, results indicate that PSWQ score and estradiol are not significantly related to R - Span Score. O - Span t ask . Results of the full model for O - Span score are presented in Table 9 and estimates for the intercept, estradiol, PSWQ score and their interaction are depic ted in Table B 11 . No variables of interest emerged as significant predictors of O - Span score. Results do not support that O - Span scores are related to PSWQ score and estradiol. 25 D ISCUSSION To better elucidate the relationship between worry and working memory , the current study investigated relationships between worry and behavioral and neurophysiological measures of working memory across multiple tasks while also taking ovarian hormones into account in women across the menstrual cycle . On the N - back task , it was hypothesized that worry would relate to deficits in WM performance (either in longer RT or reduced accuracy) and a reduce d P300 amplitude when the task was most difficult (i.e. at higher loads and more difficult trial types of the N - back ) , but only when estradiol was high . For complex span tasks, it was expected that, if worry related to task - general WM, an increase in worry would be related to poorer task performance on both span tasks when estradiol was high. If worry rela ted to phonologoical/verba l task - specific working memory, deficits related to worry were expected to occur on the R - Span task only when estradiol was high. Results revealed mixed evidence for study hypotheses. Specifically, results indicate that increases in worry symptoms across t he menstrual cycle were related to reduced accuracy on 3 - back trials of the N - back when estradiol levels were low . In contrast to other hypotheses, w orry and estradiol were not found to be related to N - back reaction time, P300 amplitude, O - Span score, and R - Span score. Findings suggest that may be specific to accuracy on more difficult trials on the N - back task The marginally sign ificant interaction between PSWQ score, estradiol and trial type provided partial support for my hypotheses . Greater worry did relate to poorer task performance when the task was more difficult. However, contrary to my hypothes e s, this effect occurred only when estradiol was relatively low for the participant. Although this finding is not in line with work by Shansky and colleagues, it may be supported by the anxiety literature, which has 26 indicated estradiol can have protective effects on other relevant functions such as fear extinc tion ( e.g. Li & Graham, 2017; Montoya & Bos, 2017) . With regards to the null findings for the relationships between worry and other measures of working memory , several possibilities must be considered. First, this study is highly novel in its methodology. Most stud ies of anxiety and working memory are cross - sectional in design whereas the current study took a repeated measures longitudinal approach . Moreover , previous studies of the role of ovarian hormones in cognition and anxiety typically do not obtain daily assays of hormones from naturally - cycling women across many time points . I t is also important to note that the models utilized in the analysis are mor e complex than analyses typically utilized in cross - section al work. A larger sample will be critical for providing the opportunity for more well powered tests of what appear to be more complex associations between worry and working memory measures . This st udy aims to gather 160 participants with useable data, so power concerns wil l be reduced , and , thus , more robust tests possible in future work . Although increasing power is important for future analyses, other considerations are equally crucial for improv ing our understanding of the relationships between worry and working memory . First, the current analyses focused on associations between within - person changes in variables of interest . However, mean level s across all the visit days may also be related to WM measures , as this would reveal between - person effects e.g., an ind ividual with higher average worry might have lower working memory than a person who has lower worry on average . Future analyses could combine within - and between - person analyses to fur ther address these different effects . R ecent studies have also indicated that estradiol may have a quadratic relationship with working memory taking the & Park, 2002) . Future analyses should attempt to examine the effect of a qu adratic estradiol term 27 in the current model once adequately powered to do so. Finally , EEG frequency and advanced signal processing techniques could be considered in future research, especially given the current findings failed to reveal effects on the P30 0 . For example, theta - gamma coupling has recently been presented as a measure o f ordering during working memory during the N - back task (Lisman & Buzsáki, 2008; Rajji et al., 2016) and may provide a more precise measure of dynamic working memory function compared to time dom ain ERP measures like the P300 . The role of estradiol in the relationship between worry and working memory was of primary interest in current study . H owever, the present results have important implications for utilizing the N - back task and span tasks for examining working memory function in participants across time. An unstructured covariance structure was found to be necessary to model N - back performance and P300 amplitude through examination of the data, as well as through comparison of the selected models with more stringent assumptions of homogeneity of variance and/or covariances (i.e. compound symmetry and heterogeneous compound symmetry structures). The necessity of this structure, along with the great heterogeneity in both variances and covaria nces observed in estimates generated for the model, suggests that there is important variability related to trial type - by - load interactions that should be accounted fo r when the N - back task is completed by participants multiple times . Future research using the N - back in a repeated - measures fashion should examine residuals for trial type - by - load interactions to determine if variances and covariances are heter o geneous , as accounting for such heterogeneity if present is critical for fitting appropriate models of this task . Related to the more complex structure required to model the N - back in the current analysis, the N - back presents an additional challenge for longitudinal work through issues of reliability. Traditionally, the N - back has been utilized in neuroimaging work, and the literature 28 base evaluating its psychometric properties is small and mixed (Jaeggi, Buschkuehl, Perrig, & Meier, 2010) . Further examination of the reliability of the task utilized in the current study could lend to the literature on this comm only used working memory measure. Additi onally, analyses for N - back reaction time, N - back P300 amplitude, O - Span Score, and R - Span score all indicate d significant patterning by the order of EEG visit s suggest ing that there were, indeed , practice effects across all tasks . Such practice effects have been observed in a recent meta - analysis of working memory tasks, even when alternative forms of the tasks were utilized (Scharfen, Jansen, & Holling, 2018) . As noted by Scharfen et al. (2018), these practice effects are likely explained by alterations in testing factors unrelated to the task, such as unfamiliarity with task rules or testing anx iety, and the development of task - specific testing strategies and/or memorization of test - specific content. Importantly, it is possible that increases are due to actual gains in WM capacity itself (Scharfen et al., 2018). However, it has been argued that t his attributes flexibility to underlying cognitive abilities that is unwarranted , given difficulty training such cognitive abilities in intervention studies where improvement in underlying cognitive abilities is explicitly targeted (Lievens, Reeve, & Heggestad, 2007) . It is difficult to disentangle the potential cause of practice effects in the current study, as all explanati ons could apply to the changes across visit demonstrated in my analyses. Given the focus on worry in this study, it is crucial to consider that state worry at the time of testing could be implicated in the practice effect demonstrated, such that greater wo rry is experienced by subjects at the first visit due to unfamiliar testing environment and tasks. Incorporating measures of state worry at the time of testing could provide insight into this issue, as our daily worry measure may not have captured worry ex perienced during the task s themselves . 29 Nonetheless, the current study represents a significant step forward in better understand ing the intricate relationship between worry and working memory by examining multiple working memory measures and taking ovarian hormones into account in women across the menstrual cycle . Because women experience chronic anxiety characterized by worry more and related cognitive impairments cognitive and emotional health. The present analyses reveal a more nuanced relationship between worry and working memory in women and calls for additional focused research studies into specific groups for w hom anxiety and worry are particularly prevalent and problematic. In taking such a targeted approach , the specificity of cognitive models of anxiety and c orresponding interventions could greatly be improve d . 30 APPENDI CES 31 APPENDIX A: Primary Tables & Figures Figures Figure 1. Illustration of study data collection across the menstrual cycle. Day 0 is menstruation. For the purposes of this study, only estradiol levels will be considered. The N - back, O - Span, and R - Span tasks are completed at each of the four EEG visits. 32 Figure 2 . Subject and session loss for each dependent measure of the study. Subjects Enrolled: 85 (323 Sessions) Subjects Estradiol Assay Processed: 71 (270 Sessions) Subjects with at least 3 EEG Visits Completed: 67 (263 Sessions) Processed N - back Behavioral data: 6 5 (2 29 sessions overall) Subjects with at least 1 EEG visit day with daily PSWQ score calculated : 67 (2 41 Sessions) Subjects with at least 1 EEG visit day with valid estradiol assay result : 6 5 (2 48 Sessions) N - back EEG Recorded: 6 5 (2 29 sessions) Subjects with Useable EEG Data after Pre - Processing: 6 5 (2 22 sessions overall ) Total usable for full model analysis: 0 - back NT: 65 (221 sessions ) 0 - back T: 65 (222 sessions) 2 - back NT: 65 (221 sessions) 2 - back T: 65 (221 sessions) 2 - back L: 65 (221 sessions) 3 - back NT: 65 (221 sessions) 3 - back T: 65 (220 sessions) 3 - back L: 65 (221 sessions) PSWQ Score and Estradiol Useable : 65 (2 29 sessions) Processed R - Span Score : 6 5 (2 28 sessions) Processed O - Span Score : 6 5 (2 29 sessions) Total usable for full model analysis (Accuracy Only) : 0 - back NT: 65 (227 sessions ) 0 - back T: 65 (227 sessions) 2 - back NT: 65 (227 sessions) 2 - back T: 65 (226 sessions) 2 - back L: 65 (220 sessions) 3 - back NT: 65 (227 sessions) 3 - back T: 65 (221 sessions) 3 - back L: 65 (223 sessions) Total usable for full model analysis: 6 5 (2 28 sessions) Total usable for full model analysis: 6 5 (2 29 sessions) 33 a) b) Figure 3 . Stimulus - locked grand average waveforms at Pz for a) non - targets, b) targets, and c) lures. Stimulus presentation occurred at 0ms. -3 -1 1 3 5 7 9 11 -200 -100 0 100 200 300 400 500 600 700 800 900 1000 Voltage (mV) 0-back 2-back 3-back -3 -1 1 3 5 7 9 11 -200 -100 0 100 200 300 400 500 600 700 800 900 1000 Voltage (mV) 0-back 2-back 3-back Time Since Stimulus Presentation (ms) Time Since Stimulus Presentation (ms) 34 Figure 3 c ) -3 -1 1 3 5 7 9 11 -200 -100 0 100 200 300 400 500 600 700 800 900 1000 Voltage (mV) 2-back 3-back Time Since Stimulus Presentation (ms) 35 a) z b) Figure 4. Breaking down the 3 - way PSWQ Score x Estradiol x Load interaction identified in the multilevel model for accuracy. (a) At high levels of estradiol (+1SD), there is no significant Load 0 - back 2 - back 3 - back Load 0 - back 2 - back 3 - back 36 interactions between load and PSWQ Score. (b) At low levels of estradiol ( - 1SD), there is a PSWQ Score is signifi cantly related to accuracy for 3 - back blocks, such that increased worry predicts worse accuracy. 37 Tables Table 1: Demographic Characteristics Characteristics Statistic Age, mean (SD) in years 20.94 (1.81) Race (%) Caucasian/White 65.67 Black/African American 19.40 More than One Race 10.47 Asian 4.48 Hispanic/Latinx (%) 16.42 Sexual Orientation (%) Heterosexual 86.57 Bisexual 7.46 Gay/ Lesbian 4.48 Asexual 1.49 Gender Identity Female 98.51 Missing 1.49 Income (%) $0 - $15,000 29.9 $15,001 - $25,000 14.9 $25,001 - $35,000 3.0 $35,001 - $50,000 4.5 $50,001 - $75,000 9.0 $75,001 - $100,000 10.4 $100,011 - $200,000 20.9 More than $200,000 6.0 Missing 1.5 Financial Supported by Other(s) 61.2 38 Table 2: PSWQ score, Estradiol and Span Task Measure Means and Standard Deviations Measure Mean (SD) Minimum Maximum PSWQ Score 41.8 (14.74 ) 16 79 PSWQ Score - Centered 0 (8.75) - 25.33 23.667 Estradiol (pg/mL) 1.739 (0.779) 0.165 5.478 Estradiol - Centered (pg/mL) 0 (0.53) - 1.8 2.25 R - Span Score 41.79 (18.85 ) 0 75 O - Span Score 45.36 ( 19.16 ) 0 75 Table 3: N - back Measure Means and Standard Deviations by Load x Trial Type Interaction Load x Trial Type Level Accuracy (%) Reaction Time (ms) P300 Amp at Pz from 300 500ms (mV) 0 - back NT (SD) 98.76 (2.01) 429.42 (74.14) 2.64 (2.76) 0 - back T (SD) 89.3 1 (9.30) 501.06 (70.98) 8.72 (4.70) 2 - back NT (SD) 98.04 (2.88) 531.65 (99.59) 2.93 (2.57) 2 - back T (SD) 78.64 (16.35) 624.28 (127.85) 6.22 (4.30) 2 - back L (SD) 76.3 8 (18.07) 665.84 (147.61) 3.84 (4.90) 3 - back NT (SD) 97.6 8 (3.09) 543.81 (108.88) 2.28 (2.71) 3 - back T (SD) 66.7 6 (15.34) 664.13 (139.79) 5.34 (4.467) 3 - back L (SD) 66.44 (17.33) 640.28 (164.44) 3.25 (4.48) 39 Table 4: Test of Type 3 Fixed Effect Significance for Multilevel Model of Accuracy (%) Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number Effect Num. df Den. df F - Value p - value Trial Type 2 214 652.48 <.0001 Load 2 219 253.38 <.0001 PSWQ Score 1 195 0.34 0.56 0 Estradiol 1 198 0.00 0.984 Trial Type x Load 3 215 148.34 <.0001 PSWQ Score x Trial Type 2 215 0.46 0.63 5 PSWQ Score x Load 2 219 0.25 0.778 Estradiol x Trial Type 2 218 0.14 0.87 3 Estradiol x Load 2 224 0.91 0.40 4 PSWQ Score x Estradiol 1 201 2.82 0.09 5 Estradiol x Load x Trial Type 3 219 0.30 0.824 PSWQ Score x Load x Trial Type 3 215 0.50 0.6 80 PSWQ Score x Estradiol x Trial Type 2 213 1.29 0.27 7 PSWQ Score x Estradiol x Load 2 217 2.80 0.06 3 PSWQ Score x Estradiol x Load x Trial Type 3 212 1.62 0.18 5 EEG Visit Number 3 154 1.10 0.351 ICC = 0.007; R 2 = 0.723 40 Table 5: Test of Type 3 Fixed Effect Significance for Multilevel Model of Reaction Time Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number Effect Num. df Den. df F - Value p - value Trial Type 2 224 379.01 <.0001 Load 2 225 335.21 <.0001 PSWQ Score 1 93.3 0.00 0.963 Estradiol 1 93.9 0.48 0.49 1 Trial Type x Load 3 224 26.16 <.0001 PSWQ Score x Trial Type 2 224 0.23 0.79 1 PSWQ Score x Load 2 225 0.69 0.504 Estradiol x Trial Type 2 224 0.65 0.521 Estradiol x Load 2 225 0.01 0.99 2 PSWQ Score x Estradiol 1 117 0.34 0.560 Estradiol x Load x Trial Type 3 224 0.55 0.64 8 PSWQ Score x Load x Trial Type 3 224 1.07 0.36 5 PSWQ Score x Estradiol x Trial Type 2 224 0.34 0.71 4 PSWQ Score x Estradiol x Load 2 225 2.30 0.103 PSWQ Score x Estradiol x Load x Trial Type 3 224 0.63 0.59 9 EEG Visit Number 3 158 8.03 <.0001 ICC = 0.302, R 2 = 0.687 41 Table 6: Test of Type 3 Fixed Effect Significance for Multilevel Model of P3 at Pz from 300 to 500ms Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number Effect Num. df Den. df F - Value p - value Trial Type 2 218 189.46 <.0001 Load 2 213 54.03 <.0001 PSWQ Score 1 128 0.42 0.5 20 Estradiol 1 128 1.39 0.2 40 Trial Type x Load 3 213 38.46 <.0001 PSWQ Score x Trial Type 2 218 0.40 0.673 PSWQ Score x Load 2 213 0.00 0.995 Estradiol x Trial Type 2 217 0.13 0.87 9 Estradiol x Load 2 212 0.21 0.81 5 PSWQ Score x Estradiol 1 169 0.20 0.655 Estradiol x Load x Trial Type 3 212 0.15 0.92 7 PSWQ Score x Load x Trial Type 3 213 0.35 0.791 PSWQ Score x Estradiol x Trial Type 2 217 0.61 0.54 5 PSWQ Score x Estradiol x Load 2 212 0.14 0.8 70 PSWQ Score x Estradiol x Load x Trial Type 3 212 0.23 0.874 EEG Visit Number 3 150 13.92 <.0001 ICC = 0.29; R 2 = 0.336 42 Table 7 : T - Scores for Differences between Least Squares Means for P3 at Pz from 300 to 500ms for Each Load by Trial Type Level Load x Trial Type Levels 1 2 3 4 5 6 7 1. 0 - back non - target 2. 2 - back non - target - 1.94 3. 3 - back non - target 2.46 3.97* 4. 0 - back target - 20.49* - 22.84* - 20.65* 5. 2 - back target - 12.37* - 12.86* - 13.82* 7.58* 6. 3 - back target - 8.35* - 9.00* - 9.81* 11.46* 2.98 7. 2 - back lure - 4.08* - 2.89 - 5.19* 11.33* 6.90* 3.65* 8. 3 - back lure - 1.97 - 1.24 - 3.18* 17.96* 8.84* 6.86* 1.45 Table 8: Test of Type 3 Fixed Effect Significance for Multilevel Model of R - Span Score related to Estradiol, and PSWQ Score Controlling for EEG Visit Number Effect Num. df Den. df F - Value p - value PSWQ Score 1 158 1.80 0.18 2 Estradiol 1 110 2.55 0.11 3 PSWQ Score x Estradiol 1 178 1.72 0.19 2 EEG Visit Number 3 160 3.66 0.01 4 ICC = 0.74 ; R 2 = 0.13; Rand om Intercept Estimate = 255.16; Residual Variance Estimate= 88.68 Table 9: Test of Type 3 Fixed Effect Significance for Multilevel Model of O - Span Score related to Estradiol and PSWQ Score Controlling for EEG Visit Number Effect Num. df Den. df F - Value p - value PSWQ Score 1 159 0.45 0.503 Estradiol 1 159 2.22 0.13 9 PSWQ Score x Estradiol 1 179 1.46 0.22 9 EEG Visit Number 3 160 6.76 <0.001 ICC = 0 .74 ; R 2 = 0.06 ; Random Intercept Estimate = 265.09, Resi dual variance estimate = 95.20 43 APPENDIX B: Ancillary Tables Table B1: Residual Correlation & Variance Estimates for Multilevel Model of Accuracy (%) Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number Load x Trial Type Levels 1 2 3 4 5 6 7 8 1. 0 - back non - target 2.86 2. 0 - back target 0.379 93.80 3. 2 - back non - target 0.440 0.297 6.41 4. 2 - back target 0.270 0.338 0.40 286.89 5. 2 - back lure 0.045 0.212 0.191 0.321 332.62 6. 3 - back non - target 0.202 0.285 0.521 0.392 0.209 7.87 7. 3 - back target 0.208 0.289 0.300 0.602 0.266 0.258 242.34 8. 3 - back lure - 0.107 - 0.039 0.165 0.0693 0.466 0.120 0.082 296.99 Random intercept estimate for subject = 1.131. Table B 2 : Estimates for the Intercept and Continuous Variable Coefficients from Multilevel Model of Accuracy (%) Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number Effect Estimate Standard Error Intercept 84.132 0.55 0 PSWQ Score - 0.022 0.06 2 Estradiol - 0.025 1.01 8 PSWQ x Estradiol 0.094 0.11 5 44 Table B 3: Least Squares Means for Accuracy for Each Load by Trial Type Level Load x Trial Type Levels Least Squares Mean (%) Standard Error 0 - back non - target 98.75 2 0 .173 0 - back target 89.250 0.588 2 - back non - target 98.019 0.207 2 - back target 78.25 5 1 .029 2 - back lure 75.938 1.303 3 - back non - target 97.65 8 0 .217 3 - back target 66.170 0.985 3 - back lure 66.145 1 .087 Table B 4: Residual Correlation & Variance Estimates for Multilevel Model of Reaction Time (ms) Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number Load x Trial Type Levels 1 2 3 4 5 6 7 8 1. 0 - back non - target 1348.8 2. 0 - back target 0.401 2051.3 3. 2 - back non - target 0.27 0 - 0.05 0 4299.1 4. 2 - back target 0.205 0.226 0.46 2 11454 5. 2 - back lure 0.120 0.033 2 0.562 0.443 15576 6. 3 - back non - target 0.217 - 0.0 90 0.7 40 0.30 6 0.209 6424.8 7. 3 - back target - 0.041 0.029 0.42 5 0.598 0.416 0.466 14143 8. 3 - back lure - 0.026 - 0.10 4 0.546 0.340 0 .533 0.647 0.520 19824 Random intercept estimate for subject = 4056.99. 45 Table B 5 : Estimates for the Intercept and Continuous Variable Coefficients from Multilevel Model of Reaction Time (ms) Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number Effect Estimate Standard Error Intercept 579.98 8.956 PSWQ Score 0.172 0.483 Estradiol - 6.306 7.992 PSWQ x Estradiol - 0.783 0.964 Table B 6: Least Squares Means for Reaction Time for Each Load by Trial Type Level Load x Trial Type Levels Least Squares Mean (ms) Standard Error 0 - back non - target 427.82 8.286 0 - back target 498.47 8.46 8 2 - back non - target 531.89 9.02 9 2 - back target 628.26 10.61 9 2 - back lure 673.22 11.435 3 - back non - target 545.04 9.52 9 3 - back target 664.27 11.15 1 3 - back lure 645.93 12.219 46 Table B7: Residual Correlation & Variance Estimates for Multilevel Model of of P3 at Pz from 300 to 500ms Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number Load x Trial Type Levels 1 2 3 4 5 6 7 8 1. 0 - back non - target 2.970 2. 0 - back target - 0.266 13.339 3. 2 - back non - target 0.052 0.145 2.689 4. 2 - back target - 0.122 0.128 0.190 14.227 5. 2 - back lure 0.200 - 0.241 0.015 0.230 19.975 6. 3 - back non - target 0.230 - 0.336 0.038 - 0.020 0.179 3.660 7. 3 - back target - 0.333 0.345 0.203 0.360 - 0.043 - 0.143 15.915 8. 3 - back lure - 0.279 0.309 0.219 0.184 0.009 - 0.113 0.366 16.246 Random intercept estimate for subject = 4.539 Table B 8 : Estimates for the Intercept and Continuous Variable Coefficients from Multilevel Model of P3 at Pz from 300 to 500ms Related to Trial Type, Load, Estradiol, and PSWQ Score Controlling for EEG Visit Number Effect Estimate Standard Error Intercept 4.31 0.289 PSWQ Score <0.001 0.014 Estradiol - 0.27 0.221 PSWQ x Estradiol 0.02 0.028 47 Table B 9: Least Squares Means for P3 at Pz from 300 to 500ms for Each Load by Trial Type Level Load x Trial Type Levels Least Squares Mean (mV) Standard Error 0 - back non - target 2.49 0.289 0 - back target 8.59 0.361 2 - back non - target 2.79 0.287 2 - back target 6.09 0.367 2 - back lure 3.70 0.400 3 - back non - target 2.13 0.295 3 - back target 5.21 0.377 3 - back lure 3.12 0.379 Table B 10 : Estimates for the Intercept and Continuous Variable Coefficients from Multilevel Model of R - Span Score related to Estradiol and PSWQ Score Controlling for EEG Visit Number Effect Estimate Standard Error Intercept 41.91 2.086 PSWQ Score 1.64 1.220 Estradiol - 0.03 0.073 PSWQ x Estradiol 0.23 0.174 48 Table B 11 : Estimates for the Intercept and Continuous Variable Coefficients from Multilevel Model of R - Span Score related to Estradiol and PSWQ Score Controlling for EEG Visit Number Effect Estimate Standard Error Intercept 41.91 2.086 PSWQ Score 1.64 1.220 Estradiol - 0.03 0.073 PSWQ x Estradiol 0.23 0.174 49 REFERENCES 50 REFERENCES Aiken, L. S., West, S. G., & Reno, R. R. (1991). Multiple Regression: Testing a nd Interpreting Interactions . Sage. American Psychiatric Association. (2013). Diagnostic a nd Statistical Manual o f Mental Disorders (DSM - 5®) . American Psychiatric Pub. Baddeley, A. (1996). The fractionation of working memory. Proceedings of the National Academy of Sciences , 93 (24), 13468 13472. Baddeley, A. D. (1986). Working memo ry (Vol. 11). Elsevier. Bailey, K., Mlynarczyk, G., & West, R. (2016). Slow wave activity related to working memory maintenance in the N - back task. Journal of Psychophysiology , 30 (4), 141. Borkovec, T. D., Ray, W. J., & Stober, J. (1998). Worry: A cognitive phenomenon intimately linked to affective, physiological, and interpersonal behavioral processes. Cognitive Therapy and Research , 22 (6), 561 576. Bredemeier, K., & Berenbaum, H. (2013). Cross - sectional and longitudinal relations between working memory performance and worry. Journal of Experimental Psychopathology , 4 (4), 420 434. Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verba l Learning and Verbal Behavior , 19 (4), 450 466. Eysenck, M. W., & Calvo, M. G. (1992). Anxiety and performance: The processing efficiency theory. Cognition & Emotion , 6 (6), 409 434. Eysenck, M. W., & Derakshan, N. (2011). New perspectives in attentional control theory. Personality and Individual Differences , 50 (7), 955 960. Eysenck, M. W., Derakshan, N., Santos, R., & Calvo, M. G. (2007). Anxiety and cognitive performance: attentional control theory. Emotion , 7 (2), 336. Ganley, C. M., & Vasilyeva, M. (2014). The role of anxiety and working memory in gender differences in mathematics. Journal of Educational Psychology , 106 (1), 105. ). Working memory for emotional facial expressions: role of the estrogen in young women. Psychoneuroendocrinology , 33 (7), 964 972. Gratton, G., Coles, M. G., & Donchin, E. (1983). A new method for off - line removal of ocular artifact. Electroencephalograph y and Clinical Neurophysiology , 55 (4), 468 484. 51 Gray, J. R., Chabris, C. F., & Braver, T. S. (2003). Neural mechanisms of general fluid intelligence. Nature Neuroscience , 6 (3), 316 322. - Depend ent Cognitive Journal of Neuroscience , 31 (14), 5286 5293. https://doi.org/10.1523/JNEUROSCI.6394 - 10.2011 Jaeggi, S. M., Buschkuehl, M., Perrig, W. J., & Meier, B. (2010). The concurrent validity of the N - back task as a working memory measure. Memory , 18 (4), 394 412. https://doi.org/10.1080/09658211003702171 Kane, M. J., Conway, A. R., Hambrick, D. Z., & Engl e, R. W. (2007). Variation in working memory capacity as variation in executive attention and control. Variation in Working Memory , 1 , 21 48. Kappenman, E. S., & Luck, S. J. (2011). ERP Components: The Ups and Downs of Brainwave Recordings. The Oxford Han dbook of Event - Related Potential Components , 3. Kessler, R. C., Petukhova, M., Sampson, N. A., Zaslavsky, A. M., & Wittchen, H. - U. (2012). Twelve - month and lifetime prevalence and lifetime morbid risk of anxiety and mood disorders in the United States. In ternational Journal of Methods in Psychiatric Research , 21 (3), 169 184. S. A. (2016). Differential Effects of Estrogen and Progesterone on Genetic and Environm ental Risk for Emotional Eating in Women. Clinical Psychological Science , 4 (5), 895 908. https://doi.org/10.1177/2167702616641637 Li, S. H., & Graham, B. M. (2017). Why are women so vulnerable to anxiety, trauma - related and stress - related disorders? The p otential role of sex hormones. The Lancet Psychiatry , 4 (1), 73 82. Lievens, F., Reeve, C. L., & Heggestad, E. D. (2007). An examination of psychometric bias due to retesting on cognitive ability tests in selection settings. Journal of Applied Psychology , 92 (6), 1672. Lisman, J., & Buzsáki, G. (2008). A neural coding scheme formed by the combined function of gamma and theta oscillations. Schizophrenia Bulletin , 34 (5), 974 980. Maeng, L. Y., & Milad, M. R. (2015). Sex differences in anxiety disorders: interactions between fear, stress, and gonadal hormones. Hormones and Behavior , 76 , 106 117. Man, M. S., MacMillan, I., Scott, J., & Young, A. H. (1999). Mood, neuropsychological f unction and cognitions in premenstrual dysphoric disorder. Psychological Medicine , 29 (3), 727 733. 52 McEvoy, L. K., Smith, M. E., & Gevins, A. (1998). Dynamic cortical networks of verbal and spatial working memory: effects of memory load and task practice. Cerebral Cortex (New York, NY: 1991) , 8 (7), 563 574. McLean, C. P., Asnaani, A., Litz, B. T., & Hofmann, S. G. (2011). Gender differences in anxiety disorders: prevalence, course of illness, comorbidity and burden of illness. Journal of Psychiatric Research , 45 (8), 1027 1035. Meyer, T. J., Miller, M. L., Metzger, R. L., & Borkovec, T. D. (1990). Development and validation of the penn state worry questionnaire. Behaviour Research and Therapy , 28 (6), 487 495. Montoya, E. R., & Bos, P. A. (2017). How oral contraceptives impact social - emotional behavior and brain function. Trends in Cognitive Sciences , 21 (2), 125 136. Moran, T. P. (2016). Anxiety and working memory capacity: A meta - analysis and narrative review. Psychological Bulletin , 142 (8), 831 864. http://dx.doi.org.proxy2.cl.msu.edu/10.1037/bul0000051 Nitschke, J. B., Heller, W., Imig, J. C., McDonald, R. P., & Miller, G. A. (2001). Distinguishing Dimensions of Anxiety and Depression. Cognitive Therapy and Research , 25 (1), 1 22. https://doi.org/10 .1023/A:1026485530405 Polich, J. (2012). Neuropsychology of P300. Oxford Handbook of Event - Related Potential Components , 159 188. Rajji, T. K., Zomorrodi, R., Barr, M. S., Blumberger, D. M., Mulsant, B. H., & Daskalakis, Z. J. (2016). Ordering informatio n in working memory and modulation of gamma by theta oscillations in humans. Cerebral Cortex , 27 (2), 1482 1490. Rosenberg, L., & Park, S. (2002). Verbal and spatial functions across the menstrual cycle in healthy young women. Psychoneuroendocrinology , 27 ( 7), 835 841. Scharfen, J., Jansen, K., & Holling, H. (2018). Retest effects in working memory capacity tests: A meta - analysis. Psychonomic Bulletin & Review , 1 25. Schoofs, D., Pabst, S., Brand, M., & Wolf, O. T. (2013). Working memory is differentially affected by stress in men and women. Behavioural Brain Research , 241 , 144 153. Shansky, R. M., Glavis - A. F. T. (2004). Estrogen mediates sex differences in stress - induced prefrontal cort ex dysfunction. Molecular Psychiatry , 9 (5), 531. 53 Shansky, R., Rubinow, K., Brennan, A., & Arnsten, A. F. (2006). The effects of sex and hormonal status on restraint - stress - induced working memory impairment. Behavioral and Brain Functions , 2 (1), 8. Shansky, Rebecca M., Bender, G., & Arnsten, A. F. T. (2009). Estrogen prevents norepinephrine alpha - 2a receptor reversal of stress - induced working memory impairment. Stress , 12 (5), 457 463. Shansky, Rebecca M., & Lipps, J. (2013). Stress - induced cognitive dysfunction: hormone - neurotransmitter interactions in the prefrontal cortex. Frontiers in Human Neuroscience , 7 . Shields, G. S., Sazma, M. A., & Yonelinas, A. P. (2016). The effects of acute stress on core executive functions: A meta - analysis and compari son with cortisol. Neuroscience & Biobehavioral Reviews , 68 , 651 668. https://doi.org/10.1016/j.neubiorev.2016.06.038 Stefanopoulou, E., Hirsch, C. R., Hayes, S., Adlam, A., & Coker, S. (2014). Are Attentional Control Resources Reduced by Worry in General ized Anxiety Disorder? Journal of Abnormal Psychology , 123 (2), 330 335. https://doi.org/10.1037/a0036343 Trezise, K., & Reeve, R. A. (2014). Working memory, worry, and algebraic ability. Journal of Experimental Child Psychology , 121 , 120 136. Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task dependent? Journal of Memory and Language , 28 (2), 127 154. Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle, R. W. (2005). An automated version of the operation span task. Behavior Research Met hods , 37 (3), 498 505. Watter, S., Geffen, G. M., & Geffen, L. B. (2001). The n - back as a dual - task: P300 morphology under divided attention. Psychophysiology , 38 (6), 998 1003. https://doi.org/10.1111/1469 - 8986.3860998 West, R., Bowry, R., & Krompinger, J . (2006). The effects of working memory demands on the neural correlates of prospective memory. Neuropsychologia , 44 (2), 197 207.