Decoding Age-Related Changes in Emotional Reactivity: The Importance of Distinguishing Positive and Negative Sensations

The Socioemotional Selectivity Theory (SST, Carstensen et al., 1999; Carstensen & DeLiema, 2018) posits that as individuals age, their perceived time horizon shortens, influencing their goals and motivations. Young adults, with a longer perceived lifespan, tend to prioritize acquiring new information and expanding their social networks to face future challenges. On the other hand, older adults, perceiving that they have less time left, focus on improving emotional well-being and deepening close relationships. Thus, according to SST, emotional stimuli and events gain in importance as we age and emotion regulation becomes increasingly critical (Carstensen & Turk-Charles, 1994). Emotional reactivity, i.e., the duration, valence and intensity of an individual’s emotional sensations in response to a stimulus, plays a crucial role in emotional well-being (e.g., Bojanowska & Zalewska, 2011; Catalino & Fredrickson, 2011). Therefore, it is likely that emotional reactivity evolves in parallel with emotional well-being during aging and that the effect of age on emotional reactivity can be predicted from the core assumptions of SST (Kliegel et al., 2007). However, the literature on the effects of age on emotional reactivity is mixed, particularly with regard to age-related differences in the valence and intensity of emotional responses (i.e., intensity of overall feelings, whether positive or negative) to images, with some studies showing stronger reactions and more extreme valence and arousal ratings in older adults (e.g., Smith et al., 2005), others finding this heightened reactivity limited to negative stimuli (e.g., Grühn & Scheibe, 2008), and still others reporting no significant age differences (e.g., Mikels et al., 2005). These discrepancies are likely due to differences in the way emotional reactivity measures have been collected (e.g., type of stimuli, type of scales) and possibly also from a lack of statistical power in studies with a small number of stimuli and participants. The aim of the present study is therefore to determine how age influences several facets of emotional reactivity within an experiment with high statistical power and finely controlled stimuli. Specifically, we will examine how age affects: 1) the valence of emotional sensations; 2) the intensity of emotional sensations; and 3) the complexity of emotional sensations (i.e., the co-occurrence of positive and negative sensations).

To better understand the results of previous studies investigating the influence of age on the valence and intensity of emotional responses to images, we propose to distinguish these studies according to theoretical and statistical criteria. Concerning theoretical criteria, Itkes et al. (2017) proposed to differentiate between two fundamental modes of valence: affective valence which refers to the emotional response to a stimulus, and semantic valence which refers to the cognitive evaluation of the inherent valence of a stimulus. Itkes et al (2017) showed that, when individuals are exposed to affective stimuli the first time, there is a strong correlation between ratings of affective valence and semantic valence (r = 0.97, p < .001), and both types of valence ratings are associated with electromyographic (EMG) activation in the corrugator supercilii and zygomaticus major muscle groups. However, these authors also noted that these two modes of valence obey different rules, since affective valence shows habituation when stimulus exposure is repeated (e.g., decrease in physiological measures such as facial EMG and heart rate), whereas semantic valence remains stable. Among the studies that examined the effect of age on valence ratings, we noted that a significant proportion assessed semantic rather than affective valence and thus did not directly assess emotional reactivity (e.g., M. Mather & Knight, 2005, Experiment 1; Mikkelsen et al., 2018). However, in view of the strong correlation between measures of affective and semantic valence obtained by Itkes et al. (2017) during initial exposure to affective stimuli, we have decided to include these studies in our literature review, but we will be careful to specify the type of measures collected when presenting the results, in particular whether affective or semantic valence was measured. Furthermore, in the bidimensional model of affect proposed by Russell (1980), valence and arousal are seen as two fundamental dimensions of emotional experience, represented by two independent orthogonal axes. However, this distinction is not universally accepted. Indeed, Kron et al. (2013) used unipolar scales to assess positive and negative sensations separately and demonstrated that the sum of ratings of these two types of sensations was strongly correlated with ratings of arousal (r = 0.75, p < .001). These findings were replicated and extended by Kron et al. (2015), who found: 1) a very high correlation between the sum of ratings of positive and negative sensations and ratings of arousal (r = 0.88, p < .001), and 2) a very high correlation between the sum of ratings of positive and negative sensations and ratings of intensity of emotional sensations (r = 0.98, p < .001), and 3) a very high correlation between ratings of arousal and ratings of intensity of emotional sensations (r = .92, p < .001). These correlations suggest an almost perfect proximity between these three measures, making the notions of arousal and intensity of emotional sensations interchangeable (e.g., see Vieillard et al., 2021). However, for the purposes of rigor and clarity, we will use the term arousal throughout this article to refer to measures of activation or excitement (e.g., see Self-Assessment Manikin, SAM; Bradley & Lang, 1994) and the term intensity of emotional sensations to refer to global ratings of positive and negative sensations (e.g., Kron et al., 2013).

Concerning statistical criteria, we conducted sensitivity analyses using G*Power 3.1.9.7. (Faul et al., 2007) on a list of 11 experiments that investigated the effect of age on emotional reactivity. These analyses revealed that, of the 11 experiments examined, seven provided sufficient statistical power (i.e., 0.80) to detect significant differences in affective/semantic valence between younger and older adults for a given emotional category of images (e.g., low-arousal positive images) only when the effect size exceeded d = 0.70 (Backs et al., 2005; M. Mather & Knight, 2005; Mikels et al., 2005; Smith et al., 2005; St Jacques et al., 2010; Wieser et al., 2006). This limitation primarily stemmed from the relatively small sample sizes in these studies. Given the well-established risk of false negatives (Vadillo et al., 2016) and false positives (Christley, 2010) in underpowered studies, we decided not to discuss their results in this article and to focus only on studies that possessed satisfactory power to show effects with a size less than or equal to d = 0.58, i.e., slightly above the size considered medium according to Cohen (Grühn & Scheibe, 2008; M. Mather & Knight, 2005; Mikkelsen et al., 2018; Streubel & Kunzmann, 2011) ¹. This selection criterion ensures that our theoretical framework and hypotheses are based on a more reliable and robust body of evidence, allowing for a nuanced and accurate interpretation of the age-related effects on emotional reactivity.

Integrating the results of the experiments with satisfactory power, we observed that the relationship between age and emotional valence - both affective and semantic - elicited by/associated with affective images of varying valence (i.e., positive and negative) and arousal levels (i.e., low and high) shows a nuanced pattern: Older adults rate low-arousal positive images more positively than younger adults (i.e., more positive semantic valence in older vs. younger adults; Mikkelsen et al., 2018) and also experience more positive emotional sensations when confronted with these images (i.e., more positive affective valence in older vs. younger adults; Streubel & Kunzmann, 2011). This age-related positivity effect extends to low-arousal negative images, for which older adults again report more positive semantic and affective valences than their younger counterparts (Mikkelsen et al., 2018; Streubel & Kunzmann, 2011). For high-arousal positive images, the pattern of results persists (affective valence: Grühn & Scheibe, 2008; semantic valence: M. Mather & Knight, 2005, Experiment 1), but Streubel and Kunzmann (2011) have pointed out that the age-related positivity effect obtained for these images is of lower magnitude than the age-related positivity effect obtained for low-arousal positive images. Finally, for high-arousal negative images, the results are more heterogeneous since Grühn and Scheibe (2008) found that older adults experience more negative emotional sensations than younger adults when confronted with these images, while no other study found a difference in valence ratings between younger vs. older adults for high-arousal negative images (M. Mather & Knight, 2005, Experiment 1; Streubel & Kunzmann, 2011). Overall, the age-related positivity effect observed for low-arousal images, whether positive or negative, aligns with theoretical proposals from SST that suggest a strategic shift in emotional processing with advancing age to optimize affective states (Carstensen et al., 1999; Scheibe et al., 2013). The fact that the age-related positivity effect decreases for high-arousal positive vs. lower-arousal positive images indicates that the arousal component of images may tax emotional regulation mechanisms and resources that promote positivity of affective or semantic valence in older adults (in the memory domain, see Kensinger, 2008; Laulan et al., 2022). Furthermore, for high-arousal negative images, the absence of an age-related positivity effect could stem from a dedifferentiation of cognitive abilities with age (see Salthouse, 2012). This dedifferentiation could particularly impair the modulation of emotional responses to high-arousal negative stimuli, as processing these stimuli requires substantial cognitive resources (Labouvie-Vief, 2009).

Examining the findings of satisfactorily powered studies on the effect of age on emotional arousal, a subtle pattern emerges here again in favor of the postulates of SST (Carstensen et al., 1999), although the small number of selected studies prompts us to be cautious in interpreting and generalizing these results. First, the study by Mikkelsen et al. (2018) shows that older adults experience higher levels of arousal than younger adults in response to low-arousal images, regardless of whether the images are positive or negative. Grühn & Scheibe (2008) have shown that this phenomenon of greater arousal in older vs. younger adults can be extended to high-arousal negative images, but that the trend is reversed for high-arousal positive images. Indeed, for high-arousal positive images, Grühn and Scheibe (2008) observed that older adults report lower levels of arousal than younger adults. Overall, these results are in line with the SST, which postulates that emotional stimuli are more salient in older adults (e.g. Carstensen et al., 1999). Based on this assumption, we predict that older adults will exhibit more intense emotional reactions to emotional events than younger adults (Carstensen, 1995; see Kliegel et al., 2007). Higher levels of emotional reactivity may be indicators of mental health (Bylsma, 2021) and contribute to well-being (Schaefer et al., 2013), since individuals with more intense emotions are better able to cope with challenges by adopting adaptive behavior based on their feelings (Waugh et al., 2011). Furthermore, the reduced levels of arousal in older adults vs. younger adults when exposed to high-arousal positive images may stem from the inclusion of erotic images in the study by Grühn & Scheibe (2008). Indeed, it has been shown that women (i.e. 52.8% of older participants in the Grühn & Scheibe, 2008 study) exhibit a decrease in pleasantness in response to erotic stimuli as they age (Ferrari et al., 2017), which can be explained by a decrease in the emotional significance of this type of stimuli.

In contrast to dimensional theories, which assume that emotional experiences are either positive or negative (e.g., Russell, 1980), the evaluative space model (ESM) proposes a more nuanced approach that recognizes the possibility of mixed emotional experiences (Cacioppo & Berntson, 1994; Norman et al., 2011). This model suggests that emotions of opposite valence can be felt simultaneously, introducing the concepts of positive and negative sensations as distinct, potentially co-occurring aspects of emotional experience. Positive sensations refer to feelings of pleasure, happiness, or any other pleasant feelings, while negative sensations encompass feelings of displeasure, sadness, or any other depleting feelings (e.g., see Kron et al., 2013, 2015). The co-occurrence of these emotional sensations contributes to what is termed emotional complexity, i.e. the degree to which an individual experiences a diverse range of emotions, including mixed or conflicting feelings, in response to a stimulus or situation (e.g., Hay & Diehl, 2011). In young adults, the analysis of verbal reports (Larsen et al., 2001) or the rating of bittersweet movies (Larsen & McGraw, 2011), in which participants reported feeling happy and sad at the same time, have provided evidence in favor of ESM hypotheses. However, the emotional aging literature is just beginning to explore the phenomenon of mixed emotions, with limited and sometimes conflicting evidence. According to the SST, the increased salience of both positive and negative emotional cues in older adults may increase affective complexity, resulting in a daily experience characterized by an increased co-occurrence of negative and positive emotions (Carstensen et al., 1999). In this sense, greater co-occurrence of positive and negative affect in older adults than in younger adults has been repeatedly demonstrated in daily life (e.g., Carstensen et al., 2011; Schneider & Stone, 2015). This phenomenon may have adaptive functions, as the co-occurrence of negative and positive emotions is associated with better emotional well-being (Hay & Diehl, 2011) and physical health in older adults (Hershfield et al., 2013). For example, according to the Dynamic Model of Affect, experiencing more positive emotions when faced with a negative event may support effective emotion regulation (e.g., Finan et al., 2011). Furthermore, the co-occurrence of positive and negative emotions is strongly associated with psychological adaptation to stress (e.g., Folkman & Moskowitz, 2000). To our knowledge, no study has investigated the complexity of emotions induced in younger vs. older adults via emotional images. However, Mather and Ready (2021) have recently conducted a study in which they asked younger and older adults to rate their negative and positive emotions before and after watching clips of sad movies, which were selected to induce a negative mood. The results showed that compared to younger adults, older adults reported a greater number of simultaneous negative and positive emotions (for similar results, see Hamilton & Allard, 2023). One limitation of this study is that only four different stimuli were used to induce a negative mood in participants, and it’s possible that these stimuli were differentially relevant to younger and older adults due to their focus on interpersonal loss.

The aim of this study is to determine the influence of age on emotional reactivity following the induction of emotions by negative, positive, and neutral images. Specifically, we will examine five aspects of emotional reactivity: 1) the valence of emotional sensations; 2) the positivity of emotional sensations; 3) the negativity of emotional sensations; 4) the intensity of emotional sensations; and 5) the complexity of emotional sensations. To this end, we will adopt measurement scales distinguishing positive from negative emotional sensations, an approach uncommon in the existing literature on aging, in which unidimensional scales have been predominantly used to assess emotional valence (e.g., Ferrari et al., 2017; Grühn & Scheibe, 2008; Mikkelsen et al., 2018). In line with the postulates of the SST (Carstensen et al., 1999) and the results of studies that have examined the link between emotional reactivity and well-being (e.g., Catalino & Fredrickson, 2011; Finan et al., 2011), we have drawn up a list of five hypotheses for our study, i.e., one for each facet of emotional reactivity.

Hypothesis: Compared to younger adults, older adults will report higher valence emotional sensations regardless of whether they are exposed to negative or positive images.

Rationale: According to the SST (Carstensen et al., 1999), older adults prioritize emotional satisfaction and positive experiences, which may influence their perception and emotional reactions to emotional stimuli. Older adults should report higher valence emotional sensations for both positive and negative images because the ability to respond more positively to positive (Catalino & Fredrickson, 2011) and negative (Finan et al., 2011) events is valuable to individuals’ well-being.

Hypothesis: Compared to younger adults, older adults will report higher positive sensations regardless of whether they are exposed to negative or positive images.

Rationale: Identical to the rationale concerning the effect of age on the valence of emotional sensations.

Hypothesis: Compared to younger adults, older adults will report higher negative sensations regardless of whether they are exposed to negative or positive images. However, these age-related differences will be less pronounced than those observed for positive sensations.

Rationale: According to the SST, the salience of emotional stimuli is higher in older adults than in younger adults (Carstensen & Turk-Charles, 1994), so we can assume that the overall emotional responses of older adults to emotional stimuli, including negative ones, will be higher than those of younger adults. However, given the importance of positive emotional sensations for well-being (e.g., Catalino & Fredrickson, 2011), we expect that the increase in positive sensations in older vs. younger adults will be more pronounced than the increase in negative sensations in older vs. younger adults.

Hypothesis: Compared to younger adults, older adults will report more intense emotional sensations regardless of whether they are exposed to negative or positive images.

Rationale: According to the SST, there is an increased salience of emotional stimuli in older adults compared to younger adults (Carstensen & Turk-Charles, 1994). More intense emotional sensations for positive images in older vs. younger adults could reflect the tendency to maximize immediate positive experiences during aging (Carstensen et al., 1999). We also expect more intense emotional sensations in older people for negative images, as higher levels of emotional reactivity are a sign of mental health even in the face of negative events (Schaefer et al., 2013).

Hypothesis: Compared to younger adults, older adults will report more mixed emotions, characterized by an increased co-occurrence of positive and negative emotional sensations. This effect will be present for both negative and positive images but will be greater for negative vs. positive images.

Rationale: According to the SST, there is an increased salience of emotional stimuli in older adults compared to younger adults (Carstensen & Turk-Charles, 1994). Furthermore, the co-occurrence of negative and positive emotions is associated with greater emotional well-being (Hay & Diehl, 2011). The age-related difference in the complexity of emotional sensations should be greater for negative images, as experiencing positive sensations when confronted with negative events may facilitate the implementation of effective emotion regulation strategies (Finan et al., 2011).

Following Grühn and Scheibe (2008), the statistical analyses in this article will be item-based rather than participant-based to be able to compare the emotional ratings collected from the young adults in our study to the normative values from the databases from which the images in our study were selected. Indeed, the procedure we used to collect emotional ratings differs substantially from that used in the International Affective Picture System (IAPS, Lang et al., 1997), the Nencki Affective Pictures System (NAPS, Marchewka et al., 2014) and the Emotional Picture Set (EmoPicS, Wessa et al., 2010) (see Procedure below), so it is important to assess the impact of these methodological choices on the results of our study.

Given the small number of studies that have investigated the effect of age on emotional reactivity and emotional complexity in the laboratory, it was not possible to reliably estimate the expected size of our effects of interest. We therefore decided to determine the number of stimuli and participants in our study pragmatically (Lakens, 2022), in particular based on: 1) the maximum duration of the experimental task tolerable for older adults; 2) the difficulty of accessing a population of physically and psychologically healthy older adults (see Laulan & Rimmele, 2024). Given our experience in recruiting and administering computerized experimental tasks to older adults, we determined that we could include a maximum of 300 images in our study (i.e., 100 items per condition) and target a maximum of 65 young and 65 older adults. We anticipated a higher attrition rate among older vs. younger adults, so the recruitment procedure resulted in an experimental sample of 67 younger and 72 older adults. A sensitivity analysis conducted using MorePower (version 6.0.4; Campbell & Thompson, 2012) indicated that for an ANOVA with a between-within interaction (i.e., 3 [valence: positive vs. neutral vs. negative] x 2 [age: young vs. older adults]), with an alpha of .05 and a total of 100 items per category, we are able to detect an effect of size η² = 0.05 with a power of 0.96 (i.e., the fifth root of 0.80 given that we proposed five independent hypotheses). A further sensitivity analysis showed that using a paired Student’s t, with an alpha of .05 and 67 participants per age group, we can detect an age effect (i.e, young vs. older adults) of size d_z = 0.35 with a power of 0.98 (i.e., the square root of 0.96 as we aim to test this effect primarily on positive and negative images in each of our hypotheses).

A total of 157 participants were initially recruited for this study. Participants were recruited through advertisements placed on the internet, including sites affiliated with the University of Geneva and online social networks (e.g., Facebook, LinkedIn), flyers placed in research centers (e.g., Campus Biotech) or places often frequented by older adults (e.g., at the university for seniors, senior clubs, and pharmacies), and through word of mouth. Participants were included only if: they were native or fluent French speakers; they had normal or corrected-to-normal vision (e.g., no color blindness); they had no history of major neurological (e.g., stroke, head injury) or psychiatric (e.g., depression, bipolar disorder) disorders. In addition, older participants were assessed for cognitive functioning with the French Telephone Interview for Cognitive Status Modified (F-TICS-m, Vercambre et al., 2010) to ensure that they did not suffer from dementia. In a validation study with an English version of the F-TICS-m (Welsh et al., 1993), Knopman et al. (2010) showed that a score of 27 was the optimal cut-off for distinguishing people with dementia from those with mild cognitive impairment. Thus, only older adults with a score of at least 28 points were included in our study (M = 33.53, SD = 3.89) (see Ballhausen et al., 2017). Among the individuals who participated in the study, seven younger adults were excluded from the analyses: three for not following instructions correctly, two due to task interruptions, and two because of improperly recorded data. Similarly, 11 older adults were excluded: five for not following instructions, five due to task interruptions, and one because of data recording issues. The task had to be interrupted for various reasons, such as the fire alarm going off, the room temperature being too high, or participants having to deal with emergency situations. The final experimental sample consisted of 67 young adults (M = 22.73 years, SD = 2.87, range = 18-30), and 72 older adults (M = 69.35 years, SD = 5.60, range = 60-82). The two age groups did not differ in sex distribution (younger adults: 68.66% female; older adults: 70.80% female, χ²(1) = 0.17, p = .780, ϕ = 0.02), but younger adults had more years of education than older adults (M = 15.94, SD = 2.00 vs. M = 12.15, SD = 3.92, t(137) = 26.84, p < .001, d = 1.21). Younger and older adults did not differ significantly on emotional reactivity scores (M = 23.38, SD = 17.20 vs. M = 25.77, SD = 16.28, t(137) = - 0.84, p = .402, d = 0.14) assessed with the Emotional Reactivity Scale (ERS; original version: Nock et al., 2008; French version: Lannoy et al., 2014). Similarly, young and older adults did not differ significantly on positive affect scores (M = 33.84, SD = 5.75 vs. M = 34.47, SD = 5. 98, t(137) = - 0.64, p = .524, d = 0.11) and negative affect scores (M = 21.00, SD = 6.59 vs. M = 19.71, SD = 5.25, t(137) = 1.28, p = .202, d = 0.22) assessed with the Positive and Negative Affect Schedule (PANAS; original version: Watson et al., 1988; French version: Caci & Baylé, 2007). However, younger adults had higher trait anxiety scores than older adults (M = 43.19, SD = 11.60 vs. M = 36.03, SD = 8.53, t(137) = 4.17, p < .001, d = 0.71), and higher state anxiety scores (M = 35.52, SD = 12.24 vs. M = 30.99, SD = 8.95, t(137) = 2.51, p = .013, d = 0.43) measured with the State-Trait Anxiety Inventory Form Y (STAI-Y; original version: Spielberger et al., 1983; French version: Bruchon-Schweitzer & Paulhan, 1993). Also, younger adults had marginally significantly higher stress scores than older adults (M = 9.70, SD = 2.92 vs. M = 8.79, SD = 2.79, t(137) = 1.88, p = .063, d = 0.32) measured with the Perceived Stress Scale (PSS 4 ; original version: Cohen et al., 1983; French version: Lesage et al., 2012). Finally, younger adults reported lower emotion suppression scores than older adults (M = 3.24, SD = 1.24 vs. M = 3.99, SD = 1.50, t(137) = - 3.18, p = .002, d = 0.54), whereas the two age groups did not differ on emotion reappraisal scores (M = 4.39, SD = 1.04 vs. M = 4.57, SD = 1.43, t(137) = - 0.87, p = .384, d = 0.15), both of these scores being measured with the Emotion Regulation Questionnaire (ERQ; original version: Gross & John, 2003; French version: Christophe et al., 2009). All participants gave written informed consent and were paid CHF 30 (approximately $31) for their participation. The study was approved by the ethics committee of the University of Geneva.

Images. The experimental materials consisted of 300 images whose main characteristics (means and standard deviations) are presented in Table 1. The 300 images were selected from international databases in which normative valence and arousal ratings were collected from young adults: 101 images were from the IAPS (Lang et al., 1997); 151 images were from the NAPS (Marchewka et al., 2014); 48 images were from the EmoPicS (Wessa et al., 2010). In all the above-mentioned databases, the valence and arousal of the images were rated on scales ranging from 1 to 9². Thus, to construct our material, we selected 100 negative images that had a normative valence rating between 1 and 3.5, of which 51 were from IAPS, 37 from NAPS and 12 from EmoPics; 100 neutral images that had a normative valence rating between 4 and 6, of which 22 were from IAPS, 65 from NAPS and 13 from EmoPics; 100 positive images that had a normative valence rating between 6.5 and 9, of which 37 were from IAPS, 49 from NAPS and 14 from EmoPics. A one-way ANOVA revealed that normative valence varied significantly between the three image categories, F(2, 297) = 2872.16, p < .001, η² = 0.95, and a Bonferroni post-hoc test specified that valence differed significantly between negative and neutral images (p < .001), between neutral and positive images (p < .001), and between negative and positive images (p < .001). Furthermore, the images in our study were selected to follow the U-shaped relationship between valence and arousal frequently obtained in normative studies (e.g., Lang et al., 1997). Thus, as shown by a one-way ANOVA, arousal varied significantly between the three image categories, F(2, 297) = 61.99, p < .001, η² = 0.30, and a Bonferroni post-hoc test highlighted that arousal differed significantly between negative and neutral images (p < .001), as well as between positive and neutral images (p < .001) but did not differ significantly between negative and positive images (p = 1). The average arousal of positive and negative images was relatively low (see Table 1) given that we wanted to match positive and negative images on this variable and that we did not include erotic images in our material even though they represent a significant proportion of high-arousal positive images in normative databases. In addition, we ensured that the content of the images did not vary between negative, neutral, and positive images. Thus, for each of the three image categories (negative, neutral, positive), 12 images contained animals, 29 images contained faces, 5 images contained landscapes, 8 images contained objects, and 46 images contained people. Image content information was already provided in the normative data of NAPS but was not provided in IAPS and EmoPics. In order to collect it for the latter two databases, three research assistants coded the content of 213 images from IAPS and 137 images from EmoPicS. To determine the content of the images (i.e., animal, face, landscape, object, people), the three raters followed the instructions described in the article by Marchewka et al. (2014). Only those images for which all three raters agreed on the category of content were retained for inclusion in our study. Finally, to avoid any possible confounding effects related to the physical characteristics of the images, the 100 positive, 100 negative, and 100 neutral images of the final selected stimulus set were matched according to: 1) visual complexity determined via JPEG size, F(2, 297) = 0.75, p = .747, η² = 0.002, and via entropy, F(2, 297) = 1.45, p = .236, η² = 0.10; luminance, F(2, 297) = 0.548, p = .578, η² = 0.004; contrast, F(2, 297) = 0.28, p = .758, η² = 0.002; the L* dimension of CIELAB color space (color’s lightness), F(2, 297) = 0.90, p = .410, η² = 0.01; the a* dimension of CIELAB color space (chromatic channel from red to green), F(2, 297) = 0.04, p = .966, η² = 0.00; the b* dimension of CIELAB color space (chromatic channel from blue to yellow), F(2, 297) = 0.03, p = .973, η² = 0.00. The physical properties regarding the NAPS images were obtained directly from the NAPS’ normative data provided by Marchewka et al. (2014). To extract the physical properties of the IAPS and EmoPicS images, we used the Python-based script developed by Marchewka et al. (2014).

Scales. We used five rating scales to assess sensations intensity, positive sensations, negative sensations, visual complexity and personal relevance. Each scale ranged from 0 (none) to 8 (very high). To assess sensations intensity, positive sensations, and negative sensations, we drew on the work of Itkes et al. (2017). We decided to use the three scales developed by Itkes et al. (2017) for three main reasons: 1) they assess affective valence rather than semantic valence, as is the case with the SAM (Bradley & Lang, 1994) that was used in the creation of the IAPS and EmoPics norms; 2) they were designed from a unipolar perspective of valence (e.g., Cacioppo & Berntson, 1994) and thus a distinction is made between positive and negative sensations, which is particularly relevant in the study of aging and emotions (see e.g., Laulan et al., 2020) and allows to evaluate the complexity of emotional sensations; 3) the possibility of answering 0 legitimizes not having felt emotional sensations and thus reduces the possibility of reporting semantic valence assessments when emotional sensations are low. Regarding visual complexity and personal relevance, these ratings were collected so that these factors could be controlled when using the material from this study in the development of paradigms to test the moderating effect of age on cognition-emotion interactions. Indeed, visual complexity is associated with emotional responses induced by images in young adults (e.g., Madan et al., 2018; Marin et al., 2016) and personal relevance influences the age-related positivity effect in memory (e.g., Hess et al., 2013; Tomaszczyk et al., 2008) as well as the effect of age on emotional reactivity (e.g., Kunzmann & Grühn, 2005). Data on visual complexity and personal relevance are not discussed in this manuscript.

The general rating procedure was adapted from the procedure that Marchewka et al. (2014) had used to develop the NAPS (especially concerning the presentation of the experiment and the instructions to the participants, the display of images and scales on the screen, the pause halfway through the experiment) (Fig. 1 A). Before starting the experiment, participants were given details about the content of the images (e.g., “You will see negative images such as a wounded animal”) and were informed that if they experienced significant discomfort during the session, they should report it immediately to the experimenter in order to stop the task. They were given instructions on the five dimensions they were asked to assess for each image, i.e., the intensity of the sensations they felt, the positive sensations they felt, the negative sensations they felt, the visual complexity of the images and the personal relevance of the images. More precisely, for the sensations intensity scale, participants were asked to rate the maximum value of any type of emotional feelings such as arousal, pleasure, displeasure; for the positive sensations scale, participants were asked to rate the feelings of pleasure, happiness, and/or any other pleasant feelings; for the negative sensations scale, participants were asked to rate the feelings of displeasure, sadness and/or any other depleting feelings; for the visual complexity scale, participants were asked to rate the overall intricacy of the image, considering the richness and diversity of visual elements such as the variety of colors, the number and types of objects, and the amount of fine details present in the image (see Harper et al., 2009); finally, for the personal relevance scale, participants were asked to rate the extent to which the content of an image was related to their daily life, their goals, and their motivations. Following Itkes et al. (2017), we emphasized to participants that the intensity of sensations scale should be considered an emotion detector and that they should select a value from 1 only if they had detected any emotion. Similarly, for the positive and negative sensation scales, participants were instructed to select a value from 1 only if they were certain they had experienced positive or negative sensations. The aim was to legitimize situations where no emotional sensations were present and thus reduce the possibility that participants would report semantic information rather than emotional sensations. To ensure their proper understanding of the instructions, the participants had to rephrase them to the experimenter (a paper with written instructions was also made available to the participants during the experiment if needed).

The rating phase began with 15 practice trials. Then, 300 images were presented to the participants in a pseudorandomized order so that no more than three images of the same valence were presented consecutively. The images were divided into 6 blocks of 50 images with breaks between each block (four short breaks and a mandatory 10-minute break halfway through the experiment) (Fig. 1 A). The order of presentation of the scales was fixed, and participants rated the intensity of their emotional sensations first to ensure that they were focusing on feelings rather than semantic knowledge for all affective ratings (see Kron et al., 2015). The presentation of each image was done in 3 steps (Fig. 1 B): 1) the image was displayed full-screen for 3 s; 2) the image was displayed in a smaller version on the left side of the screen while three scales were displayed simultaneously on the right side of the screen to rate sensations intensity, positive sensations, and negative sensations; 3) the image was displayed in the same manner on the left side of the screen and two scales appeared on the right side of the screen to rate visual complexity and personal relevance. Participants had 5 s to complete the rating of each scale by moving a circle with a computer mouse. In total, each image appeared for 28 s on the screen and the next image was automatically presented following the completion of all five rating scales for an image. The task was implemented in PsychoPy (Peirce et al., 2019) and the whole session lasted approximately 3 h.

Following the rating phase, participants completed a series of questionnaires to assess individual characteristics related to emotion and affect. These questionnaires were administered in a fixed order:

1. The ERS (original version: Nock et al., 2008; French version: Lannoy et al., 2014).

2. The PANAS (original version: Watson et al., 1988; French version: Caci & Baylé, 2007).

3. The STAI-Y (original version: Spielberger et al., 1983; French version: Bruchon-Schweitzer & Paulhan, 1993).

4. The PSS 4 (original version: Cohen et al., 1983; French version: Lesage et al., 2012).

5. The ERQ (original version: Gross & John, 2003; French version: Christophe et al., 2009).

These questionnaires were administered to characterize the participants in terms of emotional reactivity, positive and negative affects, trait and state anxiety, perceived stress, and emotion regulation strategies (suppression and reappraisal). The completion of all questionnaires took approximately 30 minutes. The entire experimental session, including the emotional reactivity task and questionnaires, lasted about 3.5 hours. At the end of the experiment, a video was shown to participants to positively restore their mood, which was potentially impacted by the negative images (link to video: https://youtu.be/fmClP8m3ExA).

Data analyses were performed using IBM SPSS Statistics version 29 for Windows (SPSS Inc, Chicago, IL, USA). All analyses were two-tailed and based on frequentist statistics. In addition, alpha was set at .05 for all analyses, and effect sizes (η² or Cohen’s d) were reported for all effects tested. Our data were systematically processed with Pearson correlations and analyses of variance (ANOVAs) in order to 1) compare our measures collected from young adults with those from the normative databases from which our images originated (i.e., IAPS, NAPS and EmoPics) and 2) test the effect of age on the various measures collected and computed, i.e., valence, positive sensations, negative sensations, intensity of sensations, and emotional complexity. These analyses were supplemented by Bonferroni post-hoc tests and Students’ t-tests when necessary to better account for the patterns of results obtained. Finally, we did not add covariates to our analyses because of their item-based nature, and because the differences observed between our two groups of participants (i.e., number of years of education, trait and state anxiety scores, perceived stress scores, emotion suppression and reappraisal scores) may be directly attributable to age differences between individuals. Controlling for these factors could artificially remove some of the variance attributable to our main factor of interest, thus reducing the construct validity of our measures (Miller & Chapman, 2001). A key challenge in aging research, as highlighted by Freund and Isaacowitz (2013), is that age cannot be randomly assigned. This makes it difficult to determine whether the differences observed between groups are due to age itself or to other background variables associated with these groups (e.g., socio-economic status, education level). Therefore, attempting to control for these variables may not clarify the results and could, in fact, obscure true age-related differences. However, we conducted Pearson correlation analyses between trait and state anxiety scores, perceived stress scores, and our dependent variables (i.e., intensity of emotional sensations, valence of emotional sensations, positivity and negativity of emotional sensations, complexity of emotional sensations) for each image category (i.e., negative, neutral, and positive), for all 139 participants. The results showed no significant correlation between anxiety and stress scores on the one hand and the variables related to emotional sensations on the other, for all image categories (all ps > .064). These results suggest that neither anxiety nor stress are significantly associated with our dependent variables in this context, reinforcing the idea that these variables should not be controlled for in the analyses.

To be able to compare the ratings of the emotional characteristics of the images collected in our study with the ratings from the IAPS, the NAPS and EmoPics, we transformed our data to obtain valence and arousal scores on scales from 1 to 9. To do this: 1) we computed a composite valence score from the positive and negative sensation ratings with the following formula, developed specifically for this study: [(Positive Sensations - Negative Sensations)/2] + 5); 2) for sensations intensity, we added 1 to all ratings to transform the 0 to 8 scale to a 1 to 9 scale.

In young adults, ratings of valence were highly consistent with those of the IAPS (r = 0.94, p < .001), NAPS (r = 0.96, p < .001), and EmoPics (r = 0.96, p < .001). Thus, the valence ratings of younger participants in our study on average strongly agreed with those in previous studies (r = 0.95, p < .001). Of note, however, mean valence ratings in our study (M = 4.84, SD = 1.32) were overall lower than those in the IAPS, NAPS, and EmoPics (M = 5.08, SD = 1.84), t(299) = - 5.67, p < .001, d = - 0.33, 95% CI [- 0.44, - 0.21]. This difference may stem from the emphasis in our study on reporting values greater than 1 only when participants were certain they were experiencing emotions. A one-way ANOVA showed that the difference between valence ratings obtained in our study and those from normative databases varied as a function of emotional category, F(2, 297) = 111.47, p < .001, η² = 0.43, 90% CI [0.36, 0.49]. Bonferroni post-hoc tests specified that this difference varied significantly between negative images (M = 0.27, SD = 0.53), neutral images (M = - 0.12, SD = 0.42), and positive images (M = - 0.86, SD = 0.66) (all ps < .001). Further analyses indicated that for negative images, valence ratings were higher in our study compared to normative ratings of the same images from the three databases (respectively, M = 3.65, SD = 1.17; M =3.38, SD = 1.44), t(99) = 5.10, p < . 001, d = 0.51, 95% CI [0.30, 0.72]; for neutral images, valence ratings were lower in our study compared to normative ratings of the same images from the three databases (respectively, M = 5.05, SD = 0.50; M = 5.17, SD = 0.58), t(99) = - 2.74, p = .007, d = - 0.27, 95% CI [- 0.47, - 0.07]; for positive images, valence ratings were lower in our study compared with normative ratings of the same images from the three databases (respectively, M = 5.82, SD = 1.07; M = 6.69, SD = 1.50), t(99) = - 13.10, p < .001, d = - 1.31, 95% CI [- 1.58, - 1.04].

In young adults, sensations intensity ratings were moderately consistent with arousal ratings from normative databases (r = 0.50, p < .001). Indeed, sensations intensity ratings in our study correlated weakly with arousal ratings from the IAPS (r = 0.50, p < .001); they correlated moderately with arousal ratings from the NAPS (r = 0.54, p < .001); they correlated strongly with arousal ratings from EmoPics (r = 0.89, p < .001). Thus, participants in the current study overall only mildly agreed with participants in previous studies regarding which images elicited low, medium, or high intensity sensations. We can also observe that the mean sensations intensity ratings in our study (M = 3.22, SD = 0.85) were overall considerably lower than arousal ratings from the IAPS, NAPS, and EmoPics (M = 5.10 SD = 0.98), t(299) = - 35.30, p < .001, d = - 2.04, 95% CI [- 2.24, - 1.84]. Here again, we can assume that this substantial difference could be due to the fact that, in our study, participants were instructed to select values greater than 1 only when they were absolutely certain that they had felt emotions. This is reflected in the high proportion of values between 0 and 1 on the sensations intensity scale among young adults (36.04% of the data). A one-way ANOVA showed that this difference between sensations intensity ratings obtained in our study and arousal ratings from normative databases varied as a function of emotional category, F(2, 297) = 6.09, p = .003, η² = 0.04, 90% CI [0.01, 0.08]. Bonferroni post-hoc tests specified that this difference was significantly smaller for negative images (M = - 1.62, SD = 0.94) than for positive images (M = - 2.02, SD = 0.74) (p = .006); it was significantly smaller for negative images than for neutral images (M = - 1.99, SD = 1.02) (p = .013); it did not vary significantly between positive and neutral images (p = 1). Further analyses indicated that for negative images, sensations intensity ratings were lower in our study compared to normative arousal ratings of the same images from the three databases (respectively, M = 3.85, SD = 0.69; M = 5.47, SD = 0.61), t(99) = - 17.29, p < . 001, d = -1.73, 95% CI [- 2.04, - 1.42]; for neutral images, sensations intensity ratings were lower in our study compared to normative arousal ratings of the same images from the three databases (respectively, M = 2.43, SD = 0.57; M = 4.41, SD = 1.17), t(99) = - 19.56, p < .001, d = - 1.96, 95% CI [- 2.29, - 1.62]; for positive images, sensations intensity ratings were lower in our study compared with normative arousal ratings of the same images from the three databases (respectively, M = 3.40, SD = 0.57; M = 5.42, SD = 0.65), t(99) = - 27.28, p < .001, d = - 2.72, 95% CI [- 3.14, - 2.29].

Average ratings of the valence of the sensations elicited by the images as a function of participants’ age (younger or older) and images’ valence (negative, neutral, or positive) are presented in Figure 2. Valence ratings calculated from positive and negative sensation scores were highly correlated between young and older adults (r = 0.95, p < .001). A 2 (age: young vs. older) x 3 (emotional category: negative vs. neutral vs. positive) mixed ANOVA was conducted with valence ratings as dependent variable. This analysis revealed a significant main effect of age, F(1, 297) = 131.15, p < .001, η² = 0.31, 90% confidence interval (CI) [0.24, 0.37]³, indicating that valence ratings were higher in older adults (M = 5.12, SD = 1.39) than in younger adults (M = 4.84, SD = 1.32). Therefore, while there was strong agreement between younger and older adults regarding the ranking of the images according to the emotional valence they elicited (i.e., more or less positive or negative), older adults overall reported that the images elicited sensations with higher valence. To determine the proportion of images with statistically different valence ratings by age, separate Student’s t-tests were performed for each of the 300 images. The results indicated that 44.00% of the images had a significant difference in valence ratings as a function of age (ps < .05), with 87.77% of these images exhibiting higher valence ratings in older adults than in younger adults. The two-way ANOVA between age and emotional category also showed a significant main effect of emotional category, F(2, 297) = 1317.70, p < .001, η² = 0.90, 90% CI [0.88, 0.91]. A Bonferroni post-hoc test showed that, as expected, valence ratings of positive images (M = 6.36, SD = 0.49) were higher than those of neutral images (M = 5.27, SD = 0.51) (p < .001) and negative images (M = 3.31, SD = 0.48) (p < .001), and valence ratings of neutral images were higher than those of negative images (p < .001). Finally, there was a significant interaction effect between age and emotional category, F(2, 297) = 15.04, p < .001, η² = 0.09, 90% CI [0.04, 0.14]. First, to clarify this interaction effect, and to determine how the effect of age on valence ratings varied by emotional category, we compared valence ratings between younger and older adults for each emotional category using Student’s t’s and a Bonferroni correction for multiple comparisons. We observed: 1) for negative images, higher valence ratings in older adults (M = 3.36, SD = 0.47) than in younger adults (M = 3.25; SD = 0.50), t(99) = - 2.71, p = .024, d = - 0.27, 95% CI [- 0.47, - 0.07]; 2) for neutral images, higher valence ratings in older adults (M = 5.49, SD = 0.48) than in younger adults (M = 5.05; SD = 0.45), t(99) = - 10.68, p < .001, d = - 1.07, 95% CI [- 1.31, - 0.82]; for positive images, higher valence ratings in older adults (M = 6.50; SD = 0.49) than in younger adults (M = 6.23; SD = 0.46), t(99) = - 6.70, p < .001, d = - 0.67, 95% CI [- 0.89, - 0.45].Then, we calculated difference scores of valence ratings between young and older adults and then compared these scores across the three emotional categories using Student’s t and a Bonferroni correction for multiple comparisons. These analyses revealed that the magnitude of the age-related difference in valence ratings was 1) greater for neutral images (M = 0.44, SD = 0.41) than for negative images (M = 0.12, SD = 0.43), t(198) = 5.37, p < .001, d = 0.76, 95% CI [0.47, 1.05]; 2) greater for neutral images than for positive images (M = 0.26, SD = 0.40), t(198) = - 3.03, p = .009, d = - 0.43, 95% CI [- 0.71, - 0.15]; 3) greater for positive images than for negative images, t(198) = 2.51, p = .039, d = 0.36, 95% CI [0.08, 0.64].

In summary, older adults reported sensations with higher valence than younger adults regardless of the type of images they were confronted with, but the magnitude of this phenomenon was greater for neutral images than for positive images, and greater for positive images than for negative images.

A novel aspect of our study was to dissociate ratings of positive and negative sensations. The analyses in this subsection were performed on raw scores of the positive sensations scale, i.e., on scores between 0 (none) and 8 (very high). Average ratings of the positive sensations elicited by the images as a function of participants’ age (younger or older) and images’ valence (negative, neutral, or positive) are presented in Figure 3. Positive sensations ratings were highly correlated between young and older adults (r = 0.93, p < .001). A 2 (age: young vs. older) x 3 (emotional category: negative vs. neutral vs. positive) mixed ANOVA was conducted with positive sensations ratings as dependent variable. This analysis revealed a significant main effect of age, F(1, 297) = 952.54, p < .001, η² = 0.76, 90% CI [0.73, 0.79], indicating that ratings of positive sensations were higher among older adults (M = 2.36, SD = 1.28) than among younger adults (M = 1.51, SD = 1.30). Thus, similar to the results presented previously regarding the effect of age on valence ratings, while there was strong agreement between younger and older adults regarding the ranking of images in terms of the positive sensations they elicited, older adults overall reported that images elicited more positive sensations than younger adults. Separate Student’s t-tests conducted for each image indicated that 75.00% of the images had a significant difference in the rating of positive sensations as a function of age (ps < .05), with 99.11% of these images for which positive sensations ratings were higher in older adults than in younger adults. The two-way ANOVA between age and emotional category also showed a significant main effect of emotional category, F(2, 297) = 884.55, p < .001, η² = 0.85, 90% CI [0.83, 0.87]. A Bonferroni post-hoc test showed that positive sensations ratings of positive images (M = 3.47, SD = 0.80) were higher than those of neutral images (M = 1.70, SD = 0.76) (p < .001) and negative images (M = 0.63, SD = 0.46) (p < .001), and positive sensations ratings of neutral images were higher than those of negative images (p < .001). Finally, there was a significant interaction effect between age and emotional category, F(2, 297) = 12.67, p < .001, η² = .08, 90% CI [0.03, 0.13]. First, to clarify this interaction effect, and to determine how the effect of age on positive sensations ratings varied by emotional category, we compared positive sensations ratings between younger and older adults for each emotional category using Student’s t’s and a Bonferroni correction for multiple comparisons. We found: 1) for negative images, higher positive sensations ratings in older adults (M = 1.02, SD = 0.28) than in younger adults (M = 0.25; SD = 0.22), t(99) = - 28.99, p < .001, d = - 2.90, 95% CI [- 3.35, - 2.45]; 2) for neutral images, higher positive sensations ratings in older adults (M = 2.22, SD = 0.58) than in younger adults (M = 1.18; SD = 0.53), t(99) = - 21.12, p < .001, d = - 2.11, 95% CI [- 2.45, - 1.76]; for positive images, higher positive sensations ratings in older adults (M = 3.84; SD = 0.70) than in younger adults (M = 3.11; SD = 0.70), t(99) = - 12.13, p < .001, d = - 1.21, 95% CI [- 1.47, - 0.95]. Then, we calculated difference scores of positive sensations ratings between young and older adults and then compared these scores across the three emotional categories using Student’s t and a Bonferroni correction for multiple comparisons. These analyses revealed that the magnitude of the age-related difference in ratings of positive sensations was 1) greater for neutral images (M = 1.04, SD = 0.49) than for negative images (M = 0.77, SD = 0.26), t(198) = - 4.91, p < . 001, d = - 0.69, 95% CI [- 0.98, - 0.41]; 2) greater for neutral images than for positive images (M = 0.73, SD = 0.60), t(198) = - 3.96, p < .001, d = - 0.46, 95% CI [- 0.84, - 0.28]; 3) not significantly different between positive and negative images, t(198) = - 0.52, p = 1, d = - 0.07, 95% CI [- 0.35, 0.20].

In summary, older adults reported higher positive sensations compared to younger adults regardless of the type of images they were confronted with, but the magnitude of this phenomenon was greater for neutral images than for positive and negative images, with no difference between positive and negative images.

The analyses in this subsection were performed on raw scores of the negative sensations scale, i.e., on scores between 0 (none) and 8 (very high). Average ratings of the negative sensations elicited by the images as a function of participants’ age (younger or older) and images’ valence (negative, neutral, or positive) are presented in Figure 4. Negative sensations ratings were highly correlated between young and older adults (r = 0.93, p < .001). A 2 (age: young vs. older) x 3 (emotional category: negative vs. neutral vs. positive) mixed ANOVA was conducted with negative sensations ratings as dependent variable. This analysis revealed a significant main effect of age, F(1, 297) = 81.70, p < .001, η² = 0.22, 90% CI [0.15, 0.28], indicating that ratings of negative sensations were higher among older adults (M = 2.13, SD = 1.66) than among younger adults (M = 1.82, SD = 1.54). Therefore, the results were equivalent to those obtained with positive sensations in that older participants reported stronger negative sensations than younger adults. Separate Student’s t-tests conducted for each image indicated that 32.33% of the images had a significant difference in the rating of negative sensations as a function of age (ps < .05), with 86.60% of these images for which negatives sensations ratings were higher in older adults than in younger adults. The two-way ANOVA between age and emotional category also showed a significant main effect of emotional category, F(2, 297) = 924.54, p < .001, η² = 0.86, 90% CI [0.84, 0.88]. A Bonferroni post-hoc test showed that negative sensations ratings of negatives images (M = 4.02, SD = 0.87) were higher than those of neutral images (M = 1.16, SD = 0.67) (p < .001) and positive images (M = 0.74, SD = 0.40) (p < .001), and negative sensations ratings of neutral images were higher than those of positives images (p < .001). Finally, there was a significant interaction effect between age and emotional category, F(2, 297) = 12.20, p < .001, η² = .08, 90% CI [0.03, 0.12]. First, to clarify this interaction effect, and to determine how the effect of age on negative sensations ratings varied by emotional category, we compared negative sensations ratings between younger and older adults for each emotional category using Student’s t’s and a Bonferroni correction for multiple comparisons. We observed: 1) for negative images, higher negative sensations ratings in older adults (M = 4.29, SD = 0.77) than in younger adults (M = 3.75; SD = 0.88), t(99) = - 6.83, p < .001, d = - 0.68, 95% CI [- 0.90, - 0.46]; 2) for neutral images, higher negative sensations ratings in older adults (M = 1.25, SD = 0.63) than in younger adults (M = 1.08; SD = 0.71), t(99) = - 3.20, p = .006, d = - 0.32, 95% CI [- 0.52, - 0.12]; for positive images, higher negative sensations ratings in older adults (M = 0.85; SD = 0.37) than in younger adults (M = 0.64; SD = 0.40), t(99) = - 5.89, p < .001, d = - 0.59, 95% CI [- 0.80, - 0.37]. Then, we calculated difference scores of negative sensations ratings between young and older adults and then compared these scores across the three emotional categories using Student’s t and a Bonferroni correction for multiple comparisons. These analyses revealed that the magnitude of the age-related difference in ratings of negative sensations was 1) greater for negative images (M = 0.53, SD = 0.78) than for neutral images (M = 0.17, SD = 0.52), t(198) = - 3.90, p < . 001, d = - 0.55, 95% CI [- 0.83, - 0.27]; 2) greater for negative images than for positive images (M = 0.20, SD = 0.35), t(198) = - 3.86, p < .001, d = - 0.55, 95% CI [- 0.83, - 0.26]; 3) not significantly different between positive and neutral images, t(198) = - 0.59, p = 1, d = - 0.08, 95% CI [- 0.36, 0.19].

In summary, older adults reported higher negative sensations compared to younger adults regardless of the type of images they were confronted with, but the magnitude of this phenomenon was greater for negative images than for neutral and positive images, with no difference between neutral and positive images. Also, to conclude this section, it is important to note that the magnitude of the difference between positive sensation ratings made by younger and older adults was greater than the magnitude of the difference between negative sensation ratings made by the two groups of participants, t(299) = - 10.95, p < .001, d = - 0.99, 95% CI [- 1.06, - 0.73]. Thus, compared to younger adults, older adults reported especially higher positive sensations. In addition, the proportion of images for which older adults reported higher positive sensations than younger adults was larger (74.3% of images) compared to the proportion of images for which older adults reported higher negative sensations than younger adults (28% of images).

Average ratings of the intensity of the emotional sensations elicited by the images as a function of participants’ age (younger or older) and images’ valence (negative, neutral, or positive) are presented in Figure 5. Emotional sensations intensity ratings were highly correlated between young and older adults (r = 0.80, p < .001). A 2 (age: young vs. older) x 3 (emotional category: negative vs. neutral vs. positive) mixed ANOVA was conducted with sensations intensity ratings as dependent variable. This analysis revealed a significant main effect of age, F(1, 297) = 1341.67, p < .001, η² = 0.82, 90% CI [0.79, 0.84], indicating that sensations intensity ratings were higher in older adults (M = 3.35, SD = 0.84) than in younger adults (M = 2.23, SD = 0.85). Therefore, while there was strong agreement between younger and older adults regarding the ranking of the images according to the sensations intensity they elicited, older adults overall reported that the images elicited higher sensations intensity than younger adults. Separate Student’s t-tests conducted for each image indicated that 80.67% of the 300 images had a significant difference in sensations intensity ratings as a function of age (ps < .05), with 98.77% of these images exhibiting higher sensations intensity ratings in older adults than in younger adults. The two-way ANOVA between age and emotional category also showed a significant main effect of emotional category, F(2, 297) = 209.14, p < .001, η² = 0.58, 90% CI [0.53, 0.63]. A Bonferroni post-hoc test showed that sensations intensity ratings for negative images (M = 3.46, SD = 0.89) were higher than those for positive images (M = 2.94, SD = 0.73) (p < .001) and neutral images (M = 1.97, SD = 0.81) (p < .001), and that sensations intensity ratings for positive images were higher than those for neutral images (p < .001). Finally, there was no significant interaction effect between age and emotional category, F(2, 297) = 0.90, p = .401, η² = 0.01, 90% CI [0.00, 0.02]. This means that older adults reported more intense sensations for a greater proportion of images than younger adults, and the magnitude ne this effect was similar for negative, neutral and positive images.

For each image and for each participant, emotional complexity was calculated using the minimum value equation (e.g., Ersner-Hershfield et al., 2008): the emotional complexity scores correspond to the minimum value of positive and negative sensations, i.e., if for a given image a participant had rated positive sensations as 1 and negative sensations as 3, the emotional complexity score for that image and participant was 1. MIN scores are the most widely used measure for characterizing the intensity of mixed emotions in the context of aging (e.g., Hamilton & Allard, 2023; M. A. Mather & Ready, 2021) and, as meta-analytic data have shown, these scores are conservative and offer safeguards against Type I error rates (Berrios et al., 2015). Average ratings of the emotional complexity of the sensations elicited by the images as a function of participants’ age (younger or older) and images’ valence (negative, neutral, or positive) are presented in Figure 6. Emotional complexity scores were weakly correlated between young and older adults (r = 0.28, p < .001). A 2 (age: young vs. older) x 3 (emotional category: negative vs. neutral vs. positive) mixed ANOVA was conducted with emotional complexity scores as dependent variable. This analysis revealed a significant main effect of age, F(1, 297) = 2395.19, p < .001, η² = 0.89, 90% CI [0.87, 0.90], indicating that emotional complexity scores were higher among older adults (M = 1.41, SD = 0.78) than among younger adults (M = 0.50, SD = 0.20). Separate Student’s t-tests conducted for each image indicated that 66.33% of the 300 images had a significant difference in emotional complexity scores as a function of age (ps < .05), with 100% of these images exhibiting higher emotional complexity scores in older adults than in younger adults. There was also a significant main effect of emotional category, F(2, 297) = 499.64, p < .001, η² = 0.77, 90% CI [0.74, 0.80]. A Bonferroni post-hoc test showed that emotional complexity scores of negative images (M = 1.48, SD = 0.19) were higher than those of neutral images (M = 0.76, SD = 0.24) (p < .001) and positive images (M = 0.62, SD = 0.18) (p < .001), and that emotional complexity scores of neutral images were higher than those of positive images (p < .001). Finally, there was a significant interaction effect between age and emotional category, F(2, 297) = 656.97, p < .001, η² = 0.82, 90% CI [0.79, 0.84]. First, to clarify this interaction effect, and to determine how the effect of age on emotional complexity scores varied by emotional category, we compared emotional complexity scores between younger and older adults for each emotional category using Student’s t’s and a Bonferroni correction for multiple comparisons. We found: 1) for negative images, higher emotional complexity scores in older adults (M = 2.41, SD = 0.34) than in younger adults (M = 0.55; SD = 0.20), t(99) = - 45.34, p < .001, d = - 4.53, 95% CI [- 5.15, - 3.84]; 2) for neutral images, higher emotional complexity scores in older adults (M = 1.04, SD = 0.36) than in younger adults (M = 0.48; SD = 0.21), t(99) = - 17.42, p < .001, d = - 1.74, 95% CI [- 2.05, - 1.43]; for positive images, higher emotional complexity scores in older adults (M = 0.78; SD = 0.21) than in younger adults (M = 0.46; SD = 0.20), t(99) = - 15.60, p < .001, d = - 1.56, 95% CI [- 1.84, - 1.26]. Then, we calculated difference scores of emotional complexity between young and older adults and then compared these scores across the three emotional categories using Student’s t and a Bonferroni correction for multiple comparisons. These analyses revealed that the magnitude of the age-related difference in emotional complexity scores was 1) greater for negative images (M = 1.86, SD = 0.41) than for neutral images (M = 0.56, SD = 0.32), t(198) = 25.04, p < .001, d = 3.54, 95% CI [3.09, 3.98]; 2) greater for negative images than for positive images (M = 0.32, SD = 0.21), t(198) = 33.48, p < .001, d = 4.73, 95% CI [4.19, 5.26]; 3) greater for neutral images than for positive images, t(198) = 6.17, p = .015, d = 0.87, 95% CI [0.58, 1.16].

In summary, older adults reported more complex emotions than younger adults regardless of the type of images they were confronted with, but the magnitude of this phenomenon was greater for negative images than for neutral images, and it was also greater for neutral images than for positive images. This means that older adults were more likely to experience complex emotions than younger adults during the experimental task, and they experienced in particular more positive sensations in conjunction with negative sensations when confronted with negative images compared to younger adults.

The aim of this study was to determine the influence of age on emotional reactivity following emotion induction by negative, positive, and neutral images, specifically by examining five aspects of emotional reactivity: 1) valence of emotional sensations; 2) positivity of emotional sensations; 3) negativity of emotional sensations; 4) intensity of emotional sensations; 5) complexity of emotional sensations. To do so, we used self-reported measures that specifically target affective valence and distinguish between positive vs. negative emotional sensations (see Itkes et al., 2017). The obtained results validated all of our hypotheses, showing that, when faced with a range of emotional images from negative to positive valence, the emotional reactions of older vs. younger adults are more positive (e.g., Streubel & Kunzmann, 2011), more intense (e.g., Mikkelsen et al., 2018) and more complex (e.g., M. A. Mather & Ready, 2021). Interestingly, the age-related increase in positive sensations was more pronounced for neutral than for both negative and positive images, confirming previous studies that have highlighted that the age-positivity effect emerges preferentially for ambiguous stimuli (emotional reactivity: van Reekum et al., 2011; perception of facial emotion: Kellough & Knight, 2012). In addition to the age-related increase in positive sensations, our results also revealed an increase in negative sensations in older vs. younger adults across all valence categories. This increase in negative sensations was more pronounced for negative than both neutral and positive images. Interestingly, the magnitude of the age-related increase in positive sensations was higher compared to the magnitude of the age-related increase in negative sensations. Thus, using a rigorous and innovative method, we have highlighted age-related differences in emotional reactivity that can be explained by psychological mechanisms described in SST (Carstensen et al., 1999) and by physiological mechanisms that complement these psychological factors.

The literature on the effect of aging on emotional reactivity is marked by significant discrepancies, e.g., some studies have shown that older vs. younger adults report more positive semantic and affective valence emotion when confronted with low-arousal negative images (Mikkelsen et al., 2018; Streubel & Kunzmann, 2011), whereas other studies have shown no age-related differences under similar conditions (Backs et al., 2005; Ferrari et al., 2017). A crucial factor explaining the heterogeneity of the results from these previous studies is the statistical power, which was not always high enough to detect effects of small to medium size (e.g., Backs et al., 2005; Wieser et al., 2006), thus compromising the replicability of the results (Anvari & Lakens, 2018). By selecting a sample of 67 young adults and 72 older adults, and by using a stimulus set comprising 100 negative, 100 neutral, and 100 positive images, we were able to detect small to medium effect sizes with a statistical power of 0.80. Our findings align with those of previous studies that have demonstrated satisfactory statistical power and have employed experimental materials similar to ours, specifically images characterized by low to moderate arousal levels. In line with Mikkelsen et al. (2018) and Streubel and Kunzmann (2011), we have indeed shown that older vs. younger adults report more positive and intense emotional experiences when exposed to weakly or moderately arousing positive and negative images. By using unipolar scales that differentiate between positive and negative sensations, we were able to describe changes in emotional experience during aging with greater granularity than studies that used only bipolar scales. We found that aging is not only characterized by the experience of more positive emotional sensations, but it is also accompanied by the experience of more negative sensations, albeit to a lesser extent. This nuanced view of emotional experience in older adults suggests a broader spectrum of emotional reactivity, which may contribute to the increased emotional complexity observed in this group. Indeed, our approach to assessing emotional reactivity also revealed an increase in emotional complexity in older vs. younger adults, extending previous similar findings with films to images (M. A. Mather & Ready, 2021).

The overall results of our study can be interpreted within the theoretical framework proposed by the SST (Carstensen et al., 1999) and can be seen as a manifestation of an adaptive strategy developed in older adults to promote a rewarding and nuanced emotional experience despite the physical and cognitive decline that accompanies aging (see Carstensen et al., 2011; Hicks et al., 2012). Indeed, our findings are in agreement with several SST postulates. First, older vs. younger adults reported more positive emotional sensations, and this to a greater extent compared to the increase in negative sensations. This finding reflects the SST postulate that older adults engage more in processing positive information in order to maintain or improve their emotional well-being (Reed & Carstensen, 2012). Second, older vs. younger adults reported more intense emotional sensations. This finding reflects the SST postulate that emotional stimuli are more salient in older adults in relation to their increased focus on the emotional significance of experiences (i.e., a tendency to pay more attention to emotionally charged information) (M. Mather & Carstensen, 2005). Third, older vs. younger adults reported more negative and complex emotional sensations. This finding reflects the SST postulate that aging is characterized by preserved or even enhanced ability to experience a broad and nuanced emotional spectrum (Carstensen et al., 2011). Thus, the appreciation of emotional experiences with advancing age, particularly positive ones, can coherently explain the wide range of emotional dynamics observed among older adults in our study. Consistent with Occam’s razor principle, also referred to as the principle of parsimony or simplicity, the ability of SST to comprehensively explain all the results of our study underscores the efficacy of this theoretical framework and its applicability in the field of emotional aging psychology.

The results of our study highlighting a simultaneous increase in positive and negative sensations, as well as greater emotional complexity in older vs. younger adults, can also be interpreted in terms of older adults’ relationship with time. Indeed, a central concept of SST posits that older adults seek to maximize their experience of now vs. later. States of expanded present-moment awareness and positive emotions have a reciprocal relationship according to broaden-and-build theory (Fredrickson, 2004; for evidence of a link between present-moment attention and positive affect in everyday life, see also Blancke et al. (2018). This focus on the present moment resonates deeply with the principles of mindfulness, which is characterized by non-judgmental attention directed towards present experience (e.g., see Klimecki et al., 2019). Since age has been shown to be positively correlated with dispositional mindfulness (e.g., Lehto et al., 2015) and older adults are overall more mindful than younger adults (e.g., Fountain-Zaragoza et al., 2018), mindfulness may be a potential mechanism by which changes in emotional experiences occur during aging. Indeed, after an 8-week mindfulness and compassion-based training vs. a relaxation training, older adults show 1) more positive semantic valence ratings for negative images (i.e., closer to values corresponding to neutral valence); 2) more negative semantic valence ratings for positive images (i.e., closer to values corresponding to neutral valence) (Shao et al., 2016). In addition, dispositional mindfulness has been shown to mediate the relationship between age and positive affect (Shook et al., 2021). In this regard, the results of our study showing an increase in both positive and negative sensations in older age when using unipolar rather than bipolar emotional scales reflect an increased emotional complexity which may be associated with an increase in mindfulness capacities in older age.

The robust age-related differences in emotional reactivity observed in our study can also be interpreted in regard of a dissociation between peripheral physiological responses and subjective emotional experience present in older adults but not in younger ones. Peripheral theories of emotion suggest that subjective emotional experience is derived from physiological responses to external or internal stimuli, and awareness of these physiological responses (e.g., Critchley et al., 2004; Damasio, 1999; Navqi et al., 2007). However, aging is characterized by a dissociation between peripheral physiological responses (e.g., cardiac, respiratory) and subjective experience according to the theory of maturational dualism (Mendes, 2010). Three mechanisms put forward in the Physiological Hypothesis of Emotional Aging (PHEA, MacCormak et al., 2022) have been proposed to explain a decoupling between subjective emotional reactivity and peripheral changes in older adults: 1) with age, there is an increase in physiological dysfunction (e.g., systemic inflammation, hypertension) that produces greater afferent noise from the viscera and peripheral transmission pathways, which can make the body’s signals less accurate and more difficult for the brain to interpret; 2) with age, the ability to distinctly perceive the body’s internal signals (i.e., interoception) decreases, which may induce changes in emotional experience (but see Mikkelsen et al., 2019); 3) with age, emotional processes may become more rooted in cognitive and situational processes rather than in direct sensory feedback, particularly because older adults have more consolidated representations of emotional body states due to the large variety and number of emotional events encountered throughout their lives. Thus, overall, the age-related differences observed in our study could be explained by a combination of these three physiological mechanisms, complementing the psychological factors described in the SST. In particular, age-related changes in interoception are potentially relevant to the interpretation of our findings, as it has been shown that, in older adults, the lower the interoceptive sensitivity, the more positive affect is reported (Haustein et al., 2023). However, an interpretation of our results based on a decoupling between peripheral physiological responses and subjective emotional experience in older adults should be taken with caution. Indeed, Lohani et al. (2018) observed greater concordance between self-reported measures of emotional sensations intensity and physiological measures (i.e., heart rate) in older vs. younger adults during the experience of sadness, going against the maturational dualism. In this sense, Gavazzeni et al. (2008) showed that, for older vs. younger adults, the dissociation between ratings of negative sensation and measures of electrodermal activity following the presentation of negative images is all the more reduced the lower the arousal of the images (but see also Kunzmann & Richter, 2009). In our study, the mean arousal of emotional images was relatively low (i.e., negative images: M = 5.50, SD = 0.60; positive images: M = 5.45, SD = 0.52), so the likelihood that our findings are due to a decoupling between subjective measures and physiological responses in older vs. younger adults seems reduced (but see Steenhaut et al., 2018).

It is important to note that our study has several limitations that may reduce the generalizability of the results. First, we used a cross-sectional design to study the effects of age on emotional reactivity, which makes it impossible to know whether the results are really due to differences between age groups, or whether they can be attributed to differences between cohorts. In addition, our sample included a substantially higher proportion of women vs. men, and this was comparable in both age groups. However, previous research has revealed that sex does not significantly influence emotional reactivity (e.g., Mikkelsen et al., 2018).

Second, our stimulus selection of low-to-medium arousal images to induce emotions limits the ecological validity of our study. This choice of stimuli covers only a fraction of the range of emotional experiences encountered in everyday life. For example, our negative images were more likely to elicit emotions such as disgust and sadness, rather than anger and fear, due to their low-to-medium level of arousal (see Riegel et al., 2016). Furthermore, previous research has shown that films are more effective than images at evoking emotions (e.g., Guo et al., 2015). Thus, the use of films could lead to more nuanced emotional reactivity, mirroring more accurately the dynamics of emotional reactivity encountered in a wider range of natural contexts. Nevertheless, our results are in line with research using films to induce emotions. For example, Fernández-Aguilar et al. (2018) observed, employing films and bipolar SAM scales to assess emotional reactivity, that: 1) older vs. younger adults reported more negative levels of affective valence, particularly after viewing films eliciting low-to-medium-arousal negative emotions (i.e., sadness); 2) older vs. younger adults reported higher levels of arousal following the presentation of films evoking low-to-medium arousal negative (i.e., disgust) or positive (i.e., tenderness) emotions.

Finally, given our findings of a general tendency for older adults to report higher emotional sensations than younger adults, we may wonder about the presence of a response bias underlying this phenomenon. Indeed, Likert scales may be subject to response bias, especially in favor of extreme responses (e.g., Moors et al., 2014), and these biases may vary according to individual characteristics (e.g., Hartley & MacLean, 2006). For the positive and negative sensation scales, from which the valence and complexity measures of emotional sensations were also derived, the presence of valence x age interaction effects reduces the likelihood that the observed age differences are due solely to response biases, as this suggests that older vs. younger adults’ responses vary as a function of context (i.e., depend on image valence). Regarding the emotional sensation intensity scale, for which we found no evidence of a valence x age interaction effect, we assessed the potential presence of response bias using quantile regression, which allowed us to examine the distributions of young and older adults’ responses across different intensity levels. This method helped us determine whether variations in emotional sensation intensity scores maintained their consistency across the distribution of scores. Our analyses revealed that differences in the emotional sensation intensity scores between younger and older adults did not follow a consistent pattern across quantiles (scores between 0.48 and 0.95 depending on quartile). This suggests that contrary to a systematic response bias whereby older adults would always use higher Likert scale scores, the observed variations across quartiles rather reflect a context-dependent emotional reactivity (i.e., dependent on image type beyond its valence, given the absence of a valence x age interaction effect). In other words, these results do not support the idea of a generalized response bias in older adults, but rather suggest a differentiated emotional experience that influences their ratings of emotional sensation intensity.

By adopting unipolar scales to distinctly measure positive and negative sensations, this study provided an in-depth exploration of changes in emotional reactivity during aging highlighting the richness of emotional experiences in older vs. younger adults. Compared to younger adults, older adults experienced more nuanced and intense emotional sensations, particularly when they were positive. These observations support the SST proposals. Mindfulness, with its emphasis on moment-to-moment awareness without judgment, aligns with observed complexities in older adults’ emotional experiences. A systematic examination of mindfulness abilities and their impact on the various facets of emotional reactivity holds substantial relevance within the SST framework. By delving into these dynamics, research could elucidate the mechanisms underpinning emotional well-being in the context of aging, thereby informing targeted interventions designed to enhance the quality of life across the lifespan. This analytical extension, grounded in empirical evidence, seeks not only to validate the theoretical constructs of SST but also to expand the scope of potential therapeutic applications aimed at fostering emotional resilience in older populations.

No potential conflict of interest was reported by the author(s).

This research was supported by the Swiss National Science Foundation (PCEFP1_186911).

Contributed to conception and design: PL, UR

Contributed to acquisition of data: PL

Contributed to analysis and interpretation of data: PL

Drafted and/of revised the article: PL, UR

Approved the submitted version for publication: PL, UR

We report assessments of statistical power, and describe all data exclusions, manipulations, and measures. Research materials with the data collected in our study averaged across participants is available online (https://osf.io/8hwzu/?view_only=71c28dac1f0149c88c5df26591dce1ae) and deidentified individual data are available upon reasonable request. All data were collected in 2022. The study’s design, the hypotheses and the analytic plan were not preregistered. Data analyses were conducted using IBM SPSS Statistics version 29 for Windows (SPSS Inc, Chicago, IL, USA). Figures were created using the ggplot2 package (Wickham, 2011) via R statistical software version 4.2.2 (R Core Team, 2022).

This study was conducted in the Brain and Behavior Lab (BBL; University of Geneva, Switzerland) and benefited from support of the BBL technical staff. We thank the “Gaieté et Harmonie” club and especially the support of Yves and Colette J. and Michelle L. We thank Andrea R., Emeline C., Lizzi P. R., Myriam D., Sarah B., Shadee T., and Yousra A. for help with data acquisition.