Izuma, K., Kennedy, K., Fitzjohn, A., Sedikides, C., & Shibata, K. (2018). Neural activity in the reward-related brain regions predicts implicit self-esteem: A novel validity test of psychological measures using neuroimaging. Journal of Personality and Social Psychology. Advance online publication. doi:10.1037/pspa0000114
2018, American Psychological Association. This paper is not the copy of record and may not exactly replicate the final, authoritative version of the article. Please do not copy or cite without authors permission. The final article will be available, upon publication, via its DOI: 10.1037/pspa0000114
Neural Activity in the Reward-Related Brain Regions Predicts Implicit Self-Esteem:
A Novel Validity Test of Psychological Measures Using Neuroimaging
Keise Izuma, Kate Kennedy, and Alexander Fitzjohn
University of York, UK
University of Southampton, UK
Nagoya University, Japan
We thank Ellen Kingsley and Olivia Rogerson for their assistance in data collection and the staff of the Neuroimaging Centre for their assistance in conducting the fMRI. K.S. was supported by JSPS Kakehnhi Grant Number 17H04789. Corresponding author: Keise Izuma, Department of Psychology, University of York, Heslington, York, YO10 5DD, UK; Tel: +44 (0)1904 323167; Email: firstname.lastname@example.org
Self-esteem, arguably the most important attitudes an individual possesses, has been a premier research topic in psychology for more than a century. Following a surge of interest in implicit attitude measures in the 90s, researchers have tried to assess self-esteem implicitly in order to circumvent the influence of biases inherent in explicit measures. However, the validity of implicit self-esteem measures remains elusive. Critical tests are often inconclusive, as the validity of such measures is examined in the backdrop of imperfect behavioral measures. To overcome this serious limitation, we tested the neural validity of the most widely used implicit self-esteem measure, the implicit association test (IAT). Given (1) the conceptualization of self-esteem as attitude toward the self, and (2) neuroscience findings that the reward-related brain regions represent an individual’s attitude or preference for an object when viewing its image, individual differences in implicit self-esteem should be associated with neural signals in the reward-related regions during passive-viewing of self-face (the most obvious representation of the self). Using multi-voxel pattern analyses (MVPA) on functional magnetic resonance imaging (fMRI) data, we demonstrated that the neural signals in the reward-related regions were robustly associated with implicit (but not explicit) self-esteem, thus providing unique evidence for the neural validity of the self-esteem IAT. In addition, both implicit and explicit self-esteem were related, although differently, to neural signals in regions involved in self-processing. Our finding highlights the utility of neuroscience methods in addressing fundamental psychological questions and providing unique insights into important psychological constructs.
Neural Activity in the Reward-Related Brain Regions Predicts Implicit Self-Esteem:
A Novel Validity Test of Psychological Measures Using Neuroimaging
In the past two decades, implicit attitude measures (most prominently, the Implicit Association Test [IAT]; Greenwald, McGhee, & Schwartz, 1998) have attracted a surge of interest from scientists and the public as a tool to uncover unconscious attitudes, that is, attitudes that an individual is unable or unwilling to report. Still, some remain skeptical of implicit measures’ validity (Blanton, Jaccard, Christie, & Gonzales, 2007; Blanton et al., 2009). Among all attitude domains to which implicit measures have been applied, no domain has attracted more skepticism than self-esteem. Implicit methods to measure self-esteem have been criticized as lacking sufficient validity (i.e., low convergent and predictive validity, low test-retest reliability) (Bar-Anan & Nosek, 2014; Bosson, Swann, & Pennebaker, 2000; Buhrmester, Blanton, & Swann, 2011; Falk & Heine, 2015; Falk, Heine, Takemura, Zhang, & Hsu, 2015; Rudolph, Schroder-Abe, Schutz, Gregg, & Sedikides, 2008), and some authors have even concluded in favor of invalidity (Buhrmester et al., 2011; Falk et al., 2015).
It is difficult, however, to make a definitive contribution to that debate, because validity has been assessed in reference to other imperfect behavioral measures. For example, Falk et al. (2015) collected nine implicit measures of self-esteem from three groups of participants (Euro-Canadians, Asian-Canadians, Japanese). The implicit measures were uncorrelated with each other across all three groups, demonstrating the low convergent validity of implicit self-esteem measures. However, we cannot conclude from these results that all implicit self-esteem measures are invalid: even if one measure was perfectly reliable and valid, no correlation would emerge in the case in which all other measures were invalid.
Similarly, the low predictive validity of implicit self-esteem measures found in prior research may be due to biases in selecting criterion variables. Researchers have typically selected criterion variables based on understanding of what explicit self-esteem is (Bosson et al., 2000; Falk et al., 2015). As a consequence, almost all criterion variables have been strongly correlated with explicit self-esteem measures, but not with implicit self-esteem measures (Bosson et al., 2000; Falk et al., 2015; for a review, see Buhrmester et al., 2011). Given the divergent validity of implicit and explicit self-esteem (Bosson et al., 2000; Buhrmester et al., 2011; Falk et al., 2015; Greenwald & Farnham, 2000; Rudolph et al., 2008), this literature may not be a fair test of the predictive validity of implicit self-esteem measures. Stated otherwise, lack of predictive validity may simply reflect unclarities in the definition of implicit self-esteem.
We aim to overcome this methodological and conceptual limitation and provide independent evidence for the validity of an implicit self-esteem measure. In particular, we investigate whether implicit self-esteem, as measured by the IAT, is associated with robust neural representations. We focused on the IAT, because it is more reliable than other implicit measures in terms of internal consistency and test-retest reliability (Bosson et al., 2000; Krause, Back, Egloff, & Schmukle, 2011; Rudolph et al., 2008). We emphasize that, although we use a neuroimaging method, our primary objective is to address a psychological question (i.e., the validity of an implicit self-esteem measure) rather than a neuroscience question (e.g., neural correlates of implicit self-esteem). We thus adopt a neuroimaging approach known as psychological hypothesis testing (Amodio, 2010).
More specifically, we test whether self-esteem IAT scores are robustly associated with neural activation in the reward-related brain regions (Bartra, McGuire, & Kable, 2013; Kolling, Behrens, Wittmann, & Rushworth, 2016; Schultz, 2015; Sescousse, Caldu, Segura, & Dreher, 2013) in response to self-face—arguably, the most obvious, immediate, and authentic representation of the self. Previous neuroimaging studies demonstrated that incidental preferences or attitudes toward various stimuli are automatically represented (i.e., without asking participants to report how they feel about the stimuli) in the reward-related areas, such as striatum and ventromedial prefrontal cortex (vmPFC) (Izuma, Shibata, Matsumoto, & Adolphs, 2017; Lebreton, Jorge, Michel, Thirion, & Pessiglione, 2009; Levy, Lazzaro, Rutledge, & Glimcher, 2011; Smith, Bernheim, Camerer, & Rangel, 2014; Tusche, Bode, & Haynes, 2010), and that individual differences in neural activities in these regions in response to rewarding stimuli are correlated with self-reported positive affect or preference for the stimuli (Bjork et al., 2004; Hariri et al., 2006; Knutson, Adams, Fong, & Hommer, 2001; Knutson, Taylor, Kaufman, Peterson, & Glover, 2005; Wu, Bossaerts, & Knutson, 2011). Furthermore, prior neuroimaging studies have shown the involvement of these reward related regions in explicit (but not implicit) self-esteem, as measured by a standardized questionnaire (i.e., trait self-esteem) (Chavez & Heatherton, 2015; Frewen, Lundberg, Brimson-Theberge, & Theberge, 2013; Oikawa et al., 2012) as well as momentary shift in how individuals feel about themselves (i.e., state self-esteem; Will, Rutledge, Moutoussis, & Dolan, 2017). The results of a more recent study (Chavez, Heatherton, & Wagner, 2017) also indicated that people’s tendency to view themselves in a positive manner is reflected in neural activations in the vmPFC, suggesting that, like preferences for consumer goods, positive attitudes toward the self are associated with activity in reward-related brain regions. In other words, neural responses in the reward-related brain regions while viewing self-face is an appropriate criterion variable, because of a close theoretical fit between what the self-esteem IAT scores and the neural responses should reflect (i.e., automatic evaluation of the self).
Thus, given that self-esteem is often conceptualized as attitude toward the self (Sedikides & Gregg, 2003), and implicit self-esteem is commonly defined as the association of the concept of self with positive or negative valence (Greenwald et al., 2002), if the IAT is a valid measure of self-esteem, its scores should be associated with neural signals in the reward-related brain regions. Stated otherwise, if self-esteem IAT scores did not reflect individual differences in any meaningful psychological trait (Buhrmester et al., 2011; Falk et al., 2015), it would be highly unlikely to observe an association between self-esteem IAT scores and neural signals in the reward-related brain regions.
In doing so, we employed a functional neuroimaging technique (functional magnetic resonance imaging or fMRI) combined with a machine learning technique called multi-voxel pattern analysis (MVPA; Haynes & Rees, 2006; Norman, Polyn, Detre, & Haxby, 2006). MVPA is known to be more sensitive in detecting different psychological, cognitive, or perceptual states than conventional fMRI data analysis (Izuma et al., 2017; Jimura & Poldrack, 2012; Sapountzis, Schluppeck, Bowtell, & Peirce, 2010) and thus suitable for identifying potentially complex associations between implicit self-esteem and neural signals in reward-related brain regions (see Methods for more details). Indeed, using MVPA, a previous fMRI study (Ahn et al., 2014) demonstrated that it is possible to predict individual differences in attitudes (political ideology) based on brain activities. Ahn et al. (2014) found that a correlation between actual political attitudes measured by a questionnaire and predicted attitudes based on MVPA was fairly high (r = 0.82), suggesting that MVPA is a reliable method for identifying the relation between an attitude measure and brain activities.
We scanned the brains of 43 individuals via fMRI while presenting them with pictures of their own face (Figure 1; see Methods for power analysis). We instructed participants to carry out a simple attention task while viewing pictures; we did not ask them to consider how they felt about themselves. Following the fMRI scanning, each participant completed the self-esteem IAT (Greenwald & Farnham, 2000) as well as two explicit self-esteem measures: (1) Rosenberg Self-Esteem Scale (RSES; Rosenberg, 1965) and (2) State Self-Esteem Scale (SSES; Heatherton & Polivy, 1991). By applying MVPA to the fMRI data, we were able to test whether participants’ level of implicit self-esteem was reliably predicted from neural signals obtained while viewing their own faces. We further examined whether explicit self-esteem scores (RSES) can be similarly predicted by neural signals in the reward-related brain regions, aiming to provide evidence for the divergent validity of implicit versus explicit self-esteem.
We recruited 48 women from the Neuroimaging Centre participant pool. All participants were current students at the University of XXX. The literature suggests gender differences in self-esteem (Bleidorn et al., 2016; Kling, Hyde, Showers, & Buswell, 1999) as well as in the relationship between perceived self-face attractiveness and self-esteem (Pliner, Chaiken, & Flett, 1990). Thus, while passive viewing of self-face would induce neural signals related to automatic evaluation of the self in both genders, the sensitivity of such responses might differ across genders. Accordingly, we recruited only females in an effort to bypass such differences in this first, validation study. Other inclusion criteria were: (1) ages of 18 to 28, (2) right-handedness1, (3) native command of the English language, (4) no history of neurological or psychiatric illness, and (5) no metal body implants or devices. We excluded five participants from the analyses: Three of them did not complete the fMRI session (two due to a problem with an fMRI scanner, one due to her decision to withdraw), and the remaining two were identified to have a brain anomaly. The final sample consisted of 43 participants aged 18-28 years (M = 20.9, SD = 2.46). All participants provided written informed consent. Ethics approval was granted by the Ethics Board of University of XX.
We estimated the effect size to be r = 0.392 based on a previous investigation (Ahn et al., 2014). As in the present study, Ahn et al. (2014) attempted to predict individual difference in social attitudes on the basis of fMRI signals. They focused on political attitudes, and reported that the correlation between predicted and actual attitudes across participants (N = 83) was r = 0.82. One crucial difference between Ahn et al.’s investigation and the present study is that our behavioral measure (i.e., IAT) is likely to be noisier than their measure of political attitudes. We estimated the difference in measurement noise based on test-retest reliability. Ahn et al. (2014) reported that the test-retest reliability of political attitudes was r = 0.952, whereas the test-retest reliability of the self-esteem IAT is r = 0.455; this is the average reliability of the following five studies (weighted by number of participants): r = 0.69 (Bosson et al., 2000), r = 0.54 (Krause et al., 2011), r = 0.54 (Rudolph et al., 2008, Study 1), r = 0.52 (Greenwald & Farnham, 2000), r = 0.39 (Rudolph et al., 2008, Study 3), and r = 0.31 (Gregg & Sedikides, 2010). Based on this information, we estimated an effect size of r = 0.392 for our study. With such an effect size, a sample size of n = 39 would achieve statistical power of β = 0.2, α = 0.05 (one-tailed). In order to account for potential data loss (e.g., due to excessive head motion in the scanner), we aimed to recruit a total of 45 participants and ended up recruiting 48.
To ensure that our sample was characterized by a wide range of self-esteem, we emailed those who expressed an interest in our fMRI study, asking them to complete an online questionnaire which included the RSES. A total of 167 individuals completed the questionnaire. 129 of the 167 respondents were eligible for the fMRI experiment (e.g., female, 18-29 years-old, right-handed, native English speakers, no history of neurological or psychiatric illness, no metal in the body). The self-esteem scores of these 129 respondents were normally distributed (range = 8-30, M = 19.14, SD = 4.66). We invited them all for participation in the fMRI study, except for most of those whose self-esteem scores hovered around the mean (16-24). Of note, the self-esteem statistics (RSES score) for our final sample (n = 43) at the pre-screening stage were: range = 8-30, M = 19.88, SD = 5.39.
We employed images of participants’ own faces as experimental stimuli during the fMRI scanning (Figure 1a). For use in the self-image presentation inside an fMRI scanner, we took four photographs of each participant under uniform lighting conditions during a 15-minute session a few weeks prior to scanning with a Nikon Coolpix s9900 digital camera (1600 × 1200 pixels). Photographs were front facing passport style, with participants displaying neutral, open-eyed expressions. We also used four scrambled images of natural scenes (i.e., not self-images; Figure 1b) as emotionally-neutral control stimuli, so that all participants viewed the same scrambled images.
------------------- Insert Figure 1 about here -------------------
We selected scrambled images as control stimuli, because we considered them emotionally neutral. Given that we aimed to predict individual differences in self-esteem from neural signals, an ideal control stimulus would induce the same attitude-related activations across all participants (e.g., neutral for everyone). It could be argued that control stimuli like faces of unfamiliar individuals are more appropriate, as they have been used in prior research (Kaplan, Aziz-Zadeh, Uddin, & Iacoboni, 2008; Sugiura et al., 2000). However, this research was concerned with brain regions specific to self-faces, and thus its objective was fundamentally different from the objective of the present study. Faces of unfamiliar individuals are not suitable control stimuli in our study: There are individual differences in face attractiveness judgement (Honekopp, 2006), and facial attractiveness/trustworthiness affects neural activity in reward-related brain regions (Mende-Siedlecki, Said, & Todorov, 2013). Hence, use of unfamiliar individuals’ faces as control stimuli would likely reduce signals in which we were interested.
Furthermore, it could be argued that, because there are so many differences between self-face and scrambled images, we cannot make strong inferences based on contrasts between these conditions. There are two key differences between the present study and typical neuroimaging research. First, again, the present study does not aim to identify brain regions specific to self-face processing. Second, we used a machine learning technique (MVPA; see below for more detail) to detect activation patterns that are associated with individual differences in the automatic evaluation of the self (implicit self-esteem). Machine learning is capable of locating specific patterns that are associated with a variable of interest from big (and noisy) data (Alpaydin, 2014). As stated above, neural signals related to individual differences in the automatic evaluation of the self should be included in the contrast of the self-face versus scrambled image conditions (especially in reward-related brain regions). If so, machine learning (MVPA) should be able to locate specific information related to it and thus predict implicit self-esteem.
The study consisted of two sessions on two separate days: (1) photo session, and (2) fMRI session. On the first day, we asked participants to complete the photo session. After we gave them general instructions on the project and fMRI safety information, we took four photographs of each participant. The photo session occurred an average of 27 days prior to the fMRI experiment. We concealed the true purpose of the study (i.e., predicting self-esteem based on brain activities) by mentioning to participants that it was concerned with neural responses to social versus non-social objects.
On the second day, during fMRI scanning, participants viewed 30 blocks. These were (1) self-images blocks, (2) scrambled-image control blocks, and (3) rest (i.e., a fixation cross) blocks (10 blocks each). Presentation of each block lasted 12 sec. In each of the self-image and scrambled-image blocks, we presented 4 different images for 2 sec each in randomized order per block (inter-stimulus interval = 1 sec). Within each block, at random intervals one image darkened for 300ms, which participants were instructed to detect and respond to as quickly as possible with a right index finger button press. We asked participants to engage in this simple task inside the scanner in order to ensure that they were paying attention to the presented images. Similar low-demanding tasks have been used in studies that examined neural responses related to automatic evaluations of various stimuli (Ahn et al., 2014; Cunningham et al., 2004; Izuma et al., 2017; Smith et al., 2014). We recorded participants’ responses within a 2 sec window post-luminance change. Given that we were interested in how individual differences in implicit self-esteem are related to brain activations, we fixed the order of blocks across all participants. After the fMRI run (a total of 6 min), each participant underwent a different fMRI run, which is unrelated to the objective of the current study (and the relevant data will not be reported here).
Following fMRI scanning, we instructed participants to engage in behavioral tasks. Participants first completed a self-esteem IAT (Greenwald & Farnham, 2000). We created the IAT with Psychopy stimulus presentation software (Peirce, 2007). The IAT comprised the four following catetories: (1) Self, (2) Other, (3) Positive, and (4) Negative. The Self category included I, My, Me, Mine, and Self, whereas the Other category included they, them, their, theirs and other. In addition, the Positive category included 10 positive words (e.g., Peace, Joy, Honest), whereas the Negative category included 10 negative words (e.g., Agony, Stupid, Useless).
Following the IAT, we administered the explicit self-esteem measures of RSES and SSES. Note that the SSES consists of three sub-scales: appearance, performance, and social. The subscales assess aspects of self-esteem that are based on physical appearance, ability, and others' evaluation, respectively. Finally, participants rated the attractiveness of their face (“how attractive do you think your face is compared to average students on campus”) on a 7-point scale (1 = Least Attractive, 4 = Average, 7 = Most Attractive). Upon completion, we paid participants £16 and debriefed them.
We calculated a self-esteem IAT score for each participant using the algorithm developed by Greenwald, Nosek, and Banaji (2003). We excluded one participant from the analyses of the behavioral data obtained during the fMRI scanning (reaction time and performance in the luminance change detection task) due to malfunction of the response box. For paired t tests, following Equation 3 of Dunlap, Cortina, Vaslow, and Burke (1996), we computed the effect sizes by
d = t[2(l -r)/n]1/2
where t is the t-statistic, r is the correlation between two measures, and n is the sample size.
fMRI Data Acquisition
We used an 8 Channel head coil, GE 3T HDx Excite MRI scanner in the Neuroimaging Centre to acquire whole brain fMRI data. Participants underwent a 13 second standard localizer scan and 12 second ASSET calibration for parallel imaging. We also obtained high resolution T1-structural scans (TE = 3 minute minimum full; TR = 7.8ms; TI = 450ms; 20° flip angle matrix = 256x256x176; FOV = 290x290x176; slice thickness = 1.13x1.13x1mm voxel size). Functional data collection consisted of a 6 min scan, gathering 120 volumes using T2*-sensitive echo-planar imaging (TE = 30ms; TR = 3000ms; 90° flip angle; matrix = 96x96; FOV = 288mm). We used horizontal orientation interleaved bottom-up acquisition, with 38 4mm slices (128x128 voxels per slice; 2mm voxel).