Keywords

1 Introduction

1.1 Background

Emotion is especially relevant to technology use, as studies of user frustration [3, 4], user satisfaction [5], user pleasure [6], and design aesthetics [7] have demonstrated. However, emotion is also complicated by individual differences, many of which stem from cultural background. This includes variation in languages, concepts, values, appropriate social behaviors, and conventions of technology use. As a consequence of such wide cultural variation, there is also cultural variation in how emotion is conceptualized, perceived, generated, and regulated. Korean hwabyeong, for example, is a culturally unique emotion concept referring to anger that develops from repeated social transgressions [8]. Emotion concepts may be technologically situated, as well. Consider the term road rage, referring to a range of frustration-induced behaviors in context of driving [9]. More recently, emotional concepts have emerged to describe emotion-laden user behavior. A gamer may rage quit after an especially frustrating experience with a computer game. While past research on user experience has not given this term its due attention, a Google search reveals about 2.9 million results for it. The prevalence of these terms raises an interesting question of how these emotional experiences compare with those of other cultures using the same technology.

Apart from these highly situated emotion terms, even generic emotion concepts (e.g., anger) show variability in what exactly they encapsulate. Despite a popular view that emotion categories are universal, evidence from neurobiology suggests that a given emotion category has no singular corresponding region or set of regions in the brain, and a given brain region has no singular corresponding emotion [2, 10]. Rather, emotion terms seem to loosely bracket a wide range of behavioral responses, which are driven by physiological and psychological processes that are not strictly emotional in and of themselves. Some of these processes bear a cognitive role in giving semantic interpretation to the behavioral response. According to this account, known as the constructionist view, emotion is an emergent phenomenon, arising from a cognitive, context-dependent interpretation of bodily sensory input [11].

Cultural variation has been found with emotion regulation, as well, particularly between Eastern and Western cultural groups [12, 13]. A common explanation for this finding is that the more collectivist Asian cultures encourage conformity to the group at the expense of one’s own individual experience; this entails down-regulating one’s emotions so as to cohere with the group. Meanwhile, Western-European cultures favor the opposite, with greater encouragement to attend to personal emotions without suppressing them [12, 13]. Emotion regulation is likely to be important for technologies which trigger emotions, especially frustration. For users who fail to employ emotion regulating strategies, a technology-induced emotion like frustration can be debilitating for use of the technology itself, which may account for deleterious effects of user frustration [4].

1.2 Regions of Connectivity in the Emotional Brain

The “emotional brain” refers to regions of the brain responsible for generating and regulating emotion, including areas of the limbic system and prefrontal cortex [14]. For this study, five regions of interest (ROIs) were selected to represent the emotional brain in the context of the experimental task, which involved rating emotional content of fear-laden and neutral images. The first of these regions is the amygdala, chosen for its well-established involvement in processing fear [15]. Cultural differences of emotion regulation, which include conceptualization and regulation, should manifest as differences in cognition. For this reason, three cognitive regions were chosen as ROIs: the anterior cingulate cortex (ACC), the ventromedial prefrontal cortex (vmPFC), and the dorsolateral prefrontal cortex (dlPFC). The ACC has been established as a mediator between emotion and cognition [16]. The vmPFC is neuroanatomically connected with the amygdala [17] and has been implicated in regulation of negative emotion and stress [18, 19]. The dlPFC is most strongly associated with cognitive control [20] and has received mixed support regarding its role in emotion regulation. Though it has no strong anatomical connection to the amygdala [21], previous studies have found cultural differences in its activation during emotion regulation tasks [1, 12]. Lindquist and colleagues [2] suggested that this region is important for the conceptualization of emotion but acknowledge that its role may also be regulatory.

The fifth ROI was the primary visual cortex (V1), the bottom-most level of hierarchy in this task-specific network. In addition to its role in bottom-up processing, Padmala and Pessoa [22] found that V1 was subject to top-down modulation in response to emotional stimuli. However, higher-order visual cortical areas are more frequently found to be the target of modulation [23,24,25]. For this reason, regulation differences may manifest as connectivity differences to V1, in addition to connectivity differences to the amygdala.

Altogether, our ROI set comprised one limbic region (amygdala), one perceptual region (V1), and three frontal regions with heterogeneous involvement in cognition and emotion processing. Given neuroimaging evidence of cultural differences in the emotion regulation, we hypothesized that dynamic causal modeling (DCM) would further show differences the in the effective connectivity among these ROIs, with Chinese participants showing a greater top-down regulation of the amygdala than US participants, as well as greater integration among frontal regions. To better operationalize these hypotheses, we explain the DCM method in the next section.

1.3 DCM and Effective Connectivity

Introduced by Friston and colleagues [26], dynamic causal modeling (DCM) is a method in which plausible networks (or models) of effective connectivity (i.e., causal influence) are created and tested for fit with existing fMRI or EEG data. Models are hypothesis-motivated node-link representations of brain networks, with nodes representing ROIs and directed links representing directed effective connectivity between the ROIs. In this way, models are deterministic, nonlinear input-output systems, with inputs being experimental stimuli and outputs being the measurements of brain activity [26].

Bayesian model selection (BMS) determines which of the models best explains the data. This involves an expectation-maximization algorithm to determine a log-evidence value for each model. The model with the highest log-evidence value is the winning model [26]. For comparison of individual connections, Bayesian model averaging (BMA) determines effective connectivity strengths for each directed connection within the winning model. These values represent the change in effective connectivity in response to experimental stimulus perturbation; for this study, this perturbation reflects differences between fear image responses and neutral image responses. Higher magnitudes indicate stronger connectivity, while positive and negative sign indicates excitation and inhibition, respectively [27].

To operationalize our hypotheses in the language of DCM, we created eight plausible models of effective connectivity (Fig. 1), comprising the ROIs discussed in the previous section. The models were manipulated primarily in terms of how the frontal regions (dlPFC, vmPFC, ACC) were connected with the remaining network.

Fig. 1.
figure 1

Plausible models of the fronto-limbic amygdala network in the context of the fear appraisal task. Amy = amygdala, dlPFC = dorsolateral prefrontal cortex, vmPFC = ventromedial prefrontal cortex, ACC = anterior cingulate cortex, V1 = primary visual cortex.

We hypothesized that measures of effective connectivity (winning model and connectivity strengths) would corroborate the individualist-collectivist hypothesis. This entailed two predictions:

  1. 1.

    Connectivity strengths from frontal regions (ACC, vmPFC, dlPFC) to the amygdala will be inhibitory and stronger for Chinese participants than for US participants.

  2. 2.

    Chinese participants will favor a network with more connections among frontal regions, reflecting greater top-down regulation.

2 Method

2.1 Participants

Ten Chinese (5 female, mean age = 23.7 years, SD = 2.1) and ten US (5 female, mean age = 21.1 years, SD = 0.54) participants were recruited for this study. All participants were English-proficient, right-handed, had normal or corrected-to-normal vision, and had no pre-existing neurological disorder. For the Chinese group, all participants reported having lived in mainland China for at least 18 years. The study was approved by the university’s Institutional Review Board, and all participants provided consent before completing the study.

2.2 Experiment

An image set was compiled comprising 36 fearful and 36 neutral images, obtained from the International Affective Picture System and categorized according to normative ratings of valence and arousal [28]. Fear images were significantly different in valence (F = 169.51, p < 0.001) and arousal (F = 494.42, p < 0.001) from neutral images, with fear images being negative in valence and high in arousal. After being fitted to the EEG cap, participants completed a practice rating trial and then five iterations of main trials. For each image, a trial began with a 4-s fixation period, followed by a 4-s presentation of the image, and then a prompt in which participants reported their level of emotion on a 5-point scale.

2.3 Data Collection and Analysis

EEG data was recorded using a 62-electrode EEG cap (Electro-Cap International, Inc.) with a modified 10–20 arrangement [29], reference to the left ear lobe, and grounding between Afz and Fpz. For preprocessing, EEG signals were bandpass filtered to a range of 0.01–75 Hz, subjected to artifact subspace reconstruction [30] and channel interpolation, and then re-referenced to an average reference.

After data collection and cleaning, Bayesian model selection was applied to the dataset using the eight models in Fig. 1. For each group, Bayesian model averaging was then applied the group’s winning model to obtain connectivity values for each directed connection. To keep our analysis manageable, we examined only connections that were at least 0.1 in strength magnitude and at least 95% in posterior probability. To test for predicted cultural differences in effective connectivity strengths, one-tailed t-tests were conducted on all six directed connections to the left and right amygdala, with a Bonferroni-adjusted criterion of significance.

3 Results

3.1 Winning Models

Bayesian model selection revealed that Chinese participants (CH) favored a model bearing a fully integrated dlPFC (Fig. 2, Model #5), with connections to all regions, while US participants favored a model wherein the dlPFC was connected with only the amygdala and V1 (Model #3b). The only differences between these models are the dlPFC-ACC and dlPFC-vmPFC connections, which are present only in Model #5. Using significance criterion of ΔF > 3 [31], both winning models significantly exceed their second-place models in log-evidence (ΔF = 346 for Chinese group, ΔF = 138 for US group).

Apart from winning models, the Chinese group’s second-place model was 3b—the US group’s winning model. Meanwhile, US participants’ winning model was followed by model 3a. Between 3b and 3a, the critical difference is that 3b possesses ACC-vmPFC connections, while 3a does not; both of these models treat the dlPFC as working in parallel to the other two frontal regions.

Fig. 2.
figure 2

Log-evidence of models for US and Chinese participants. Asterisks indicate winning models. ΔF indicates difference in log-evidence between winning model and second-place model for each group.

3.2 Connectivity Gains

Bayesian model analysis was applied to the two winning models to obtain connectivity strength values (Fig. 3). Of the connections in both group’s winning models, 11 of the 36 connections for the Chinese group and 7 of the 28 connections for the US group passed the strength threshold. All connections the passed this threshold were significant, indicated by the posterior probabilities. For connections to the amygdala, three of these were strong for the Chinese group (r-ACC-amygdala = −0.203, l-vmPFC-amygdala = −0.101, r-vmPFC-amygdala = −0.116), and none were strong for the US group. One-direction t-test revealed significance differences for r-ACC-amygdala (t(18) = −6.55, p < 0.00001) and l-vmPFC-amygdala (t(18) = −2.74, p = 0.0067). For both connections, Chinese participants showed negative connectivity gain while US participants to positive gain.

Despite differences in winning models, BMA did not find strong connections from the dlPFC to other regions for either group, but it did reveal a strong r-ACC-dlPFC connection for Chinese participants (−0.257). In addition to this culture-unique connection, Chinese participants also showed especially strong r-ACC connections to r-amygdala (−0.203) and r-V1 (.193), while the US group showed only a strong connection to r-V1 (−0.125). The Chinese group showed strong inhibitory connections from three frontal regions to their ipsilateral amygdala regions, while the US group showed no strong connections to the amygdala.

Fig. 3.
figure 3

Connectivity gains and posterior probabilities (parenthesis) for the winning model of Chinese (CH) and US groups. Left and right graphs refer to left and right hemispheres, respectively. Solid and dotted lines refer to forward and backward connections, respectively.

4 Discussion

4.1 Summary of Findings

Our first hypothesis was that Chinese participants would demonstrate greater top-down regulation of the amygdala, and this prediction was borne out in connectivity strengths. Chinese participants showed greater overall inhibition of the amygdala by frontal regions (r-ACC, l-vmPFC, r-vmPFC) than did US participants; meanwhile, US participants showed only one significant connection to the amygdala, and this was excitatory. This suggests that as part of the emotion appraisal task, which involved no explicit instruction to employ emotion regulation strategies, Chinese participants more readily employed cognitive regions to downregulate amygdala activity. This is consistent with past findings of cultural differences in emotion regulation and coheres well with the explanation that participants conform to the values of their respective cultures [1, 12]. That is, while Eastern culture encourages curtailment of personal emotion, Western culture encourages personal experience.

Our second hypothesis was that frontal regions would show more connections among each other for Chinese participants than for US participants. Consistent with this, Chinese participants showed a winning model bearing more connections with the dlPFC, suggesting that for Chinese participants the dlPFC had an especially important interactions with the vmPFC and ACC. Regarding these connections, the only one that passed the strength threshold was r-ACC-dlPFC for Chinese participants. No connections from the dlPFC passed the strength threshold or showed a remarkable cultural difference, suggesting that the dlPFC’s role did not involve regulating other ROIs. While the dlPFC consistently showed little effect on the remainder of the network, effect on the dlPFC was uniquely strong for Chinese participants. Additionally, Chinese participants overall showed greater connectivity from the ACC than US participants. Given these effective connectivity differences, and given that the ACC is an established intermediary between emotional and cognitive processes [16], it appears that Chinese participants more readily employed cognitive processes in their appraisal of the images; this includes conceptualization, which is an important facet of the constructionist account of emotion [2].

Altogether, these findings suggest that for an emotion appraisal task, Chinese participants showed greater involvement of cognitive regions. Part of this involvement suggests cultural differences in the downregulation of emotional response, while another part (ACC-dlPFC connectivity) suggests something other than emotion regulation, possibly emotion conceptualization.

4.2 Implications for User Experience

What might these cultural differences entail for usability? Although the emotional content of this study was not technologically situated, our findings of cultural differences in emotion processing give some reason for speculating into emotional facets of usability and design. If cultural background influences how an individual responds to an emotion-evoking stimulus, then the consequences of emotion-laden aspects in a system, such as user frustration, may also depend on cultural background. User frustration is an especially relevant example. Like fear, it is also salient, negative, and characterized by a strong physiological response [32]. A user’s resilience against technology-induced frustration may well depend on their ability and tendency to downregulate their emotional response, whether implicitly or through explicit strategies such as reappraisal. Future studies should thus investigate how such culture-based differences of emotion may manifest themselves in human-computer interaction.

5 Conclusion

An obvious limitation of this study is its scope, in that fear was the sole emotion category selected. However, given that fear has a more concise etiology compared to other emotions, a finding of cultural difference in fear processing suggests that cultural differences pervade other, more neurologically complex emotion categories as well. It also bears mentioning that the emotional aspect of the experimental stimuli was established merely with the content of an image and did not emerge from technology use. Consequently, this study does not inform cultural differences with respect to technology-induced emotions such as frustration and pleasure, despite our earlier speculations. Additional research is needed to establish how culture may affect a user’s emotional response to usability issues or hedonic design. Still, our findings do establish one instance of cultural differences in the connectivity of the emotional brain, which gives good reason to anticipate cultural differences in technologically situated emotions.

DCM itself bears technical limitations as a method, including the assumption of prior probabilities and the assumption of deterministic input and output. The latter is regarded as the more severe of the two, but these limitations are commonly accepted as part of DCM [26].

Altogether, DCM revealed that Chinese participants showed greater involvement of frontal regions of the emotional brain, including stronger regulation of the amygdala compared to US participants, as well as a more integrated dlPFC. Both findings suggest a culture-based difference in emotional appraisal of fear-laden images. In the context of cross-cultural design, these findings suggest caution in how both culture and emotion is accounted for in technology.