https://orcid.org/0000-0003-4406-0366
Abstract
Facial expressions are powerful social signals that motivate feelings and actions in the observer. Research on face processing has overwhelmingly used static facial images, which have limited ecological validity. Previous research on the age-related positivity effect and age differences in social motivation suggest that older adults might experience different evoked emotional responses to facial expressions than younger adults. Here, we introduce a new method to explore age-related differences in evoked responses to dynamic facial expressions across adulthood.
We used dynamic facial expressions which varied by expression type (happy, sad, and angry) and expression magnitude (low, medium, and full) to gather participant ratings on their evoked emotional response to these stimuli along the dimensions of valence (positive vs negative) and arousal.
As predicted, older adults rated the emotions evoked by positive facial expressions (happy) more positively than younger adults. Furthermore, older adults rated the emotion evoked by negative facial expressions (angry and sad) more negatively than younger adults. Contrary to our predictions, older adults did not differ significantly in arousal to negative expressions compared with younger adults. Across all ages, individuals rated positive expressions as more arousing than negative expressions.
The findings provide some evidence that older adults may be more sensitive to variations in dynamic facial expressions than younger adults, particularly in terms of their estimates of valence. These dynamic facial stimuli that vary in magnitude are promising for future studies of more naturalistic affect elicitation, studies of social incentive processing, and use in incentive-driven choice tasks.
Many of our interactions involve real-time social feedback in the form of facial expressions. Researchers have identified adult age differences in the ability to distinguish different facial expressions (Ruffman et al., 2008) and have noted the importance of using age-compatible stimuli in face-processing research (Ebner et al., 2010). Recent research using dynamic (rather than static) facial expressions shows that more dynamic stimuli minimize age differences in emotion recognition (Holland et al., 2019). Emotional facial expressions convey important social information, so it is important to understand the affective states evoked by facial expressions in others, and how this may vary at different stages of adult development. Critical to developing this understanding is having tools that more authentically represent the facial expressions of people individuals encounter every day.
Work on emotional contagion (Hatfield et al., 1993) suggests that viewing facial expressions in others not only elicits similar facial expressions in the viewer but also changes the emotional state of the viewer (Hatfield et al., 2014). In particular, studies have shown that stronger facial expressions can evoke stronger emotional reactions (Wild et al., 2001). Although many of these studies focused on specific emotions (i.e., surprise), the field of affective science has begun to describe affective experiences along two independent dimensions, valence and arousal (Barrett & Russell, 1998; Russell, 1980). This approach is useful for exploring adult age-related differences in emotional experience (Nielsen et al., 2008). Older adults typically experience more positive emotions (Carstensen et al., 2000, 2011, 2020), and using ratings along these dimensions allows researchers to focus on differences in valence, or positive versus negative emotional experiences.
Studies of age-related differences in socioemotional information processing have demonstrated age-related differences not only in emotional experiences but also in the evaluation of positive versus negative information, or the age-related positivity effect (Bailey et al., 2020). Although older literature suggests that this effect is due to increased attention to and memory for positive information in older age (Mather & Carstensen, 2005), more recent literature suggests that reduced arousal in older age, particularly to negative stimuli, could also contribute to the effect (Carstensen & DeLiema, 2018). Thus, it is possible that the age-related positivity effect could contribute to variations in response to the facial expressions of others, and that this could be through age differences in valence or arousal (see Supplementary Material for further discussion).
Here, we introduce a new set of incremented, dynamic facial expressions. Using a previously validated emotional facial expression database (Ebner et al., 2010; Holland et al., 2019), we created dynamic, incremented facial stimuli spanning a range of expression magnitudes. Although not commonly used in aging research, dynamic stimuli provide more ecological validity and have been shown to improve emotion recognition in older adults (Holland et al., 2019). We then evaluated the emotional response evoked by these stimuli, asking participants to rate their emotional experience in terms of valence and arousal. Based on the age-related positivity effect, we predicted an increase in positive ratings for positive facial expressions with age. We also predicted a decrease in arousal ratings for negative facial expressions with age. Based on prior work showing that stronger emotional facial expressions evoked stronger emotional responses in viewers, we also predicted there would be an increase in arousal and more extreme valence ratings (more positive ratings for positive expressions, more negative ratings for negative expressions) with increasing expression magnitude. Exploratory analyses also examined the curvilinear effects of age.
Method
Participants
Two hundred and seven participants (age: M = 49.04, SD = 16.74, and range = 21–78 years old) were recruited for an online study using Qualtrics Panels. Total survey time was approximately 30 min and was approved by the Duke University Institutional Review Board. See Supplementary Material for participant characteristics.
Affect Response Rating Task
Here, we modified the ratings used in prior studies of affective responses (Nielsen et al., 2008). Participants viewed a total of 54 videos (2 s per video, three facial expressions [happy, sad, and angry] with three magnitudes per emotion for six face models; see Figure 1). Videos included one model of each gender (male and female participants) for each age group (younger, middle-aged, and older adult); please see Supplementary Material for additional details. Stimulus presentation was randomized. Prior to beginning the task, participants were instructed about the arousal and valence scales used (Supplementary Figures 1 and 2). After viewing each video, participants rated their own emotional arousal on a 7-point Likert scale (very aroused to not at all aroused). Participants also rated the valence of their evoked emotion (very positive to very negative) on a separate, 7-point Likert scale. Ratings were self-paced.
Sample of face models and the expression magnitudes participants were shown (examples for Happy, Sad, and Angry expressions). After each video, participants provided arousal and valence ratings on a Likert scale (right). Videos were adapted from dFACES Database, which is available online at https://faces.mpdl.mpg.de/imeji/
Analysis
Participant ratings of evoked valence and arousal were coded such that higher values reflected higher arousal or more positive valence (Figure 1). Ratings were collapsed across the six face models to calculate mean ratings per participant for each emotion type and expression magnitude. Linear regression analyses were carried out using the lm() function from the base stats package in R (Version 4.2.2). Models testing the hypotheses included age, emotion type, expression magnitude, and the interaction between these terms. Model 1 includes age, emotion type, and their interactions whereas Model 2 added expression magnitude and its interactions to this model. Exploratory analyses also tested the quadratic effects of age and its interaction with experimental conditions (Models 3 and 4). Here, we report the analyses that provide the best fit to our data and use them to assess our hypotheses. Full analysis details and results, including all models tested and model comparison, can be found in Supplementary Material.
Results
Evoked Valence
Exploratory Model 4 provided the best fit to the evoked valence ratings. With this model, we found evidence supporting our first hypothesis, there would be an increase in positive ratings for positive facial expressions with age. A significant interaction between the quadratic effect of age and the positive/negative expression contrast indicated that older individuals differed more in how they rated their response to positive and negative expressions than younger individuals (Supplementary Table 3). Follow-up tests (Simonsohn, 2018; Supplementary Figure 3) suggest that this quadratic effect is driven by different patterns in ratings of positive and negative expressions. These analyses show an age-related increase in ratings in response to positive expressions that level off in older age (break-point age 60). Contrastingly, there is an age-related decrease in ratings in response to negative expressions (angry and sad) in young adulthood and middle age that flips to an age-related increase in ratings for older adults (break-point age 62).
Evoked Arousal
Exploratory Model 3 provided the best fit to the evoked arousal ratings. We predicted there would be a decrease in arousal ratings for negative facial expressions with age. Although the quadratic effect of age and positive/negative expression contrast were significant predictors of arousal ratings, there was no interaction between these terms (Supplementary Table 4). Although positive expressions were rated as more arousing than negative expressions (Figure 2B), this did not vary by age. Furthermore, follow-up tests suggest that while the quadratic effect of age is driven by an age-related decrease in arousal ratings across young adulthood, this effect shifts to an age-related increase in arousal ratings across middle age and older adulthood (break-point age 41). This is not consistent with our hypothesis.
Mean ratings for (A) arousal (1 = not at all aroused and 7 = very aroused) and (B) valence (1 = very negative, 4 = neutral, and 7 = very positive) by emotion category (angry, happy, and sad) and expression magnitude (low, medium, and full expression) by standardized age.
All Evoked Ratings
We predicted that across all ages, there would be more extreme evoked ratings with increasing expression magnitude. There was a significant interaction between the positive/negative expression contrast and both expression magnitude contrasts (Supplementary Table 3) on valence ratings. As shown in Figure 2A, the difference in ratings between positive and negative stimuli was greater for medium than low expressions and greater for full than other expression magnitudes. These results provide support for our third hypothesis. However, we found no evidence for this hypothesis in the arousal data. None of the models that included magnitude as a predictor provided a better fit than simpler models.
Discussion
Here, we introduce a new, incremented version of the existing Dynamic FACES database (Holland et al., 2019) and use these stimuli to investigate adult age differences in evoked emotional responses. We found strong support for our first hypothesis, that older adults would rate their response to positive facial expressions (happy) more positively than younger adults. Older adults also rated their response to negative facial expressions (angry and sad) more negatively than younger adults. However, we found no support for our second hypothesis, that older adults would display lower arousal to negative facial expressions. Instead, we found lower arousal to negative (vs positive) expressions in all ages. We found mixed support for the third hypothesis, that there would be stronger ratings for their response to stronger facial expressions. Although we found evidence of sensitivity to expression magnitudes in valence, we did not see corresponding sensitivity in arousal.
Our finding of an age-related increase in evoked valence ratings for positive (vs negative) expressions is consistent with the age-related positivity effect, which posits that older adults are attentive to and remember positive stimuli more than negative stimuli (Reed et al., 2014). Much of the work establishing this positivity effect used extreme-group, cohort research designs (Reed et al., 2014), limiting the ability to determine at what age this difference occurs. Our exploratory analyses show that these age-related effects are actually curvilinear, with most of the age-related differences occurring during early and middle adulthood (<60 years old), with little age-related differences in older adulthood (>62 years old), suggesting this is driven by differences in early adulthood. However, as these observations are based on exploratory analyses in cross-sectional data, hypothesis-testing research, and longitudinal studies are needed to confirm this is an actual developmental trend. Also, the greater sensitivity to the valence of emotional facial expressions in older adults observed here is consistent with the growing literature showing that older adults have stronger emotional empathy (Beadle & de la Vega, 2019; Beadle et al., 2013) and prosociality (Cutler et al., 2021; Sparrow et al., 2021) than younger adults.
We found no evidence for our prediction that older adults would be less aroused by negative (angry and sad) facial expressions (Carstensen & DeLiema 2018). Instead, exploratory analyses found a curvilinear effect of age, with arousal rating dipping in middle age. However, these findings also need to be further validated. The lack of predicted age differences in arousal is consistent with other studies of evoked emotions (Deckert et al., 2020; Nielsen et al., 2008), along with studies of emotional arousal and memory for older adults, which have found no differences in memory for positive and negative high-arousal words (Kensinger, 2008) and even memory enhancement for high-arousal words (Nashiro & Mather, 2011).
We found mixed evidence for our third hypothesis, that there would be stronger ratings for stronger facial expressions. In the valence ratings, we found evidence that people experienced stronger emotional responses to videos displaying greater expression magnitude. However, this did not translate into our arousal ratings. It is also possible that the thresholds created for low and medium magnitudes were set too high, but the variation in valence ratings for these emotions suggests that they were well-calibrated. Another possibility is that people are less sensitive to variations in the emotional intensity of negative emotions. For instance, previous research has shown that adults react to and register anger and similar negative expressions more rapidly than other emotions, likely due to the evolutionary benefits of early threat recognition (Mather & Knight, 2006).
There are a number of limitations in this study. Here, we generated dynamic images from a previously static set, demanding some manipulation of images. This manipulation could affect participants reaction to the stimuli (see Supplementary Material for examination of ratings of naturalness). It is also possible that different emotional expressions emerge along different timescales; in other words, it is possible that a “low” happy expression is stronger than a “low” angry expression. Future research is needed to attempt to equate the intensity of these different emotional expressions (c.f., Domes et al., 2008). Additionally, although the stimulus database contained an age and basic gender-diverse sample, these stimuli lack racial and cultural diversity, as all of the face models in this set are White and German adults. The creation of a more racially and culturally diverse facial expression stimuli set would be greatly beneficial to the field (Hamilton et al., 2022).
Taken together, these findings demonstrate the potential for using incremented dynamic facial expressions to elicit affective responses. Emotion valence and incrementation of stimuli by level of expression (low, medium, and full) were not only perceived by participants but elicited the expected distinct reactions. Future studies of aging should consider using more ecologically valid stimuli such as the dynamic facial expressions used here, as the use of such stimuli can minimize age differences (Holland et al., 2019) and lead to a more accurate assessment of the strengths, and vulnerabilities, of aging.
Author Notes
S. J. Abiodun is now at Princeton University. G. A. McAllister is now at Meta.
Funding
This work was supported by grants from the National Institute on Aging (Funder ID: 10.13039/100000049) (P30-AG034424 and R00-AG042596 to G. R. Samanez-Larkin; T32-AG000029 to K. L. Seaman).
Conflict of Interest
None .
Data Availability
This study was not preregistered. Project details and analysis scripts are available here: https://osf.io/3bzwt/.
