Aging and Attention to Self-Selected Emotional Content: A Novel Application of Mobile Eye Tracking to the Study of Emotion Regulation in Adulthood and Old Age

Abstract

Previous studies of attentional deployment to a single stream of experimenter-selected affective stimuli have found that compared to younger adults, older adults attend relatively more to positive and less to negative stimuli, and this can relate to better mood for them. Past studies of situation selection have yielded a contrasting picture of age similarity. In everyday life, attentional deployment is fundamentally and dynamically related to situation selection, but prior studies have investigated them only in isolation. We present new research using mobile eye tracking to test for age differences in selections of emotional stimuli and attention to self-selected choices after a negative mood induction. Younger, middle-aged, and older individuals (N=150) were either instructed to specifically try to regulate their mood state or not before having their selections, attention, and mood recorded. We used a database-oriented method to analyze fixations to positive, negative, and neutral videos once selected. Findings suggested more similarities than differences among age groups in what material was selected, how participants attended to selected material, and how their choices and attention predicted mood. Situation selection also had a more consistent relationship with mood than attentional deployment. These results suggest that age differences in attention are less apparent when participants have flexibility to avoid and choose stimuli than when viewing a predetermined fixed set of stimuli. Thus, emotion regulation strategies of selection and attention may show more age similarities when they interact than when studied in isolation.

Findings that older adults report more positive affective experience have led researchers to investigate potential age differences in the use and effectiveness of different emotion regulation strategies. This work has generally been guided by two conceptual models, one from the emotion literature and one from the aging literature. On one hand, Gross’ process model of emotion regulation (e.g., Gross, 1998) has led to theory and research on particular emotion regulation strategies that may vary in use or effectiveness by age (e.g., Shiota & Levenson, 2009; Urry & Gross, 2010). On the other hand, Carstensen’s socioemotional selectivity theory (SST; e.g., Carstensen, 2006) has generated the hypothesis that motivational changes caused by shifts in time perspective lead older adults to prioritize hedonic goals and therefore favor positive over negative or neutral stimuli in their information processing. This has led to numerous studies of potential age-related positivity effects in different types of information processing (e.g., Charles, Mather, & Carstensen, 2003; Kennedy, Mather, & Carstensen, 2004; Mather & Carstensen, 2005).

One example of work that grew out of SST is older adults’ tendency to direct attention toward positive and away from negative aspects of the environment (see Isaacowitz, 2012 for a summary). Though this pattern is consistent with SST’s notion of age-related positivity effects (Isaacowitz, Wadlinger, Goren, & Wilson, 2006), it can also be classified as the emotion regulation strategy of attentional deployment within the process model, thus locating it at the intersection of both approaches.

More recently, we turned to investigate another emotion regulation strategy specified by the process model: situation selection. This strategy comes into play even before attentional deployment, which is used within an emotion-eliciting situation; situation selection involves making choices about which potentially emotional situations someone will (or will not) expose themselves to in the first place. Conceptual models have suggested that older adults should be especially likely to utilize situation selection to avoid exposure to events that are potentially challenging to regulate (e.g., Charles, 2010). In sum, there are important theoretical reasons to investigate how both attentional deployment and situation selection are related to age.

Assessing Attentional Deployment and Situation Selection in Isolation

To test whether older adults are more positive in attentional deployment, in prior work, we used stationary eye tracking—recording participants’ gaze patterns to emotionally salient areas of interest as a fixed, experimenter-selected set of valenced stimuli were displayed on a computer screen (e.g., Isaacowitz et al., 2006). These studies generally recorded younger and older adults’ fixation patterns as they viewed emotional faces or images. In several studies, we found clear evidence that older adults are more likely to use positive looking—looking less at negative stimuli, and more at positive ones—as a regulatory strategy, and that it can be effective for them in terms of positive mood outcomes (see Isaacowitz, 2012). These patterns have also been replicated in other labs (e.g., Nikitin & Freund, 2011).

To test whether older adults are more positive in their situation selection, we designed an “Affective Environment” in which participants choose to engage with stimuli varying in valence and have their choices and mood recorded. Across several behavioral studies, findings of age differences in situation selection have been somewhat mixed (see Sands, Livingstone, & Isaacowitz, in press). The first several studies found no evidence of age differences in choices (Isaacowitz, Livingstone, Harris, & Marcotte, 2015; Rovenpor, Skogsberg, & Isaacowitz, 2013), though a more recent study in which instructions were varied found some evidence for more positive choices with age (Livingstone & Isaacowitz, 2015). Despite some evidence for potentially greater positivity in older adults’ choices, we have not found any age differences in how choices predict mood: More positive choices predict better moods, regardless of age.

Assessing Attention to Self-Selected Stimuli: Rationale and what has Been Possible So Far (Attentional Selection)

The research thus far has provided stronger evidence for age-related positivity in attentional deployment than situation selection. One problem with this conclusion, however, is that situation selection and attentional deployment have been studied separately, though they are intimately related in practice. Once a stimulus in the environment is selected for engagement, there are different ways of attending to it: For example, fixating relatively more or less toward the most affectively-laden parts of the selected stimuli. These patterns may vary by age and may also have differential impact on mood.

Studies of attentional deployment show that when confronted with stimuli that are likely to alter their mood, older adults direct their attention in ways consistent with emotion regulation (Isaacowitz, 2012); the question is whether they do this when allowed to choose their stimuli rather than having the experimenter provide the stimuli in a fixed stream for them. Perhaps older individuals never voluntarily choose to engage with negative content in the first place, rendering moot how they might fixate toward negative stimuli at all, though the situation selection research does not suggest this is the case. On the other hand, some older adults may select to interact with negative stimuli when they find it interesting or believe they have the resources to manage their emotional response to it. They may then activate a more positive looking pattern—fixating away from negative—as a form of attentional deployment to the negative stimuli they have chosen to watch. As one example, someone may choose to watch a scary film because they think it would be interesting, but then look away every time something scary seems like it may happen—the same choice with a different pattern of attention. However, we cannot determine these patterns by studying situation selection or attentional deployment in isolation.

Thus, the interplay of situation selection (choices) and attentional deployment (fixation) is of particular interest when considering potential age differences and positivity in emotion regulation. To date, technological limitations have prevented analyzing both concurrently. In stationary eye-tracking studies of attentional deployment, participants have had no choice about what images or videos appear on the screen; the fixed order of stimuli is determined by the experimenter. In affective environment studies of situation selection, generally only the choices are recorded, as individual viewings vary from person to person.

Mobile eye tracking presents a possible way to measure both situation selection and attentional deployment simultaneously, as it can record both the choice and fixation on the selected stimuli. Mobile eye tracking allows participants to be untethered from a stationary location (such as one chair facing one computer monitor), so they can move around more freely in space. This can be useful for studying many behaviors, such as visual search in space (e.g., Macdonald & Tatler, 2015). For emotion regulation, mobile eye tracking could be used to investigate behavior in ambulatory settings with multiple potential emotional inputs.

In one study, we used mobile eye tracking to track where participants looked when they selected their own stimuli, though technical limitations meant that we could only track time spent looking at selected videos, rather than fixations to specific areas of the videos themselves. We called this “attentional selection”—attention to self-selected stimuli, broadly speaking (see Table 1 for an overview of these constructs). In line with the behavioral situation selection results, the study found minimal age differences in how much younger vs. older adults looked at their positive vs. negative vs. neutral selections (Isaacowitz et al., 2015). This construct is limited, however: We could ask whether participants looked more at negative than positive selected videos, but not whether some participants choose negative but then look less at the most negative parts of the videos (a more positive looking pattern within a negative selection).

Table 1.

Summary of Previous Research on Situation Selection, Attentional Selection, and Attentional Deployment

Construct	Operational Definition	Relevant References	Previous Findings
Situation Selection: choosing to enter or avoid a situation	Assessed via affective choices in an affective environment (number of choices, percent time viewing material)	Rovenpor et al (2013), Isaacowitz et al (2015); Livingstone & Isaacowitz (2015); Sands et al. (2015)	Generally find no age differences in valence of selections or effects of selections on mood
Attentional Selection: generally attending to selected material	Assessed via percent fixation to chosen stimulus (e.g., video)	Isaacowitz et al (2015)	Found minimal age differences in attending to valenced selections
Attentional Deployment: attending to certain aspects within an environment	Assessed via percent fixation to affectively salient LookZones (areas of interest) within a standardized stimulus selected by experimenter	Isaacowitz et al (2006), Isaacowitz (2012), Nikitin & Freund (2011)	Generally find age x valence interactions in attention to negative vs. positive LookZones and parallel age-related effects on mood

Thus, while we have used mobile eye tracking in a limited way in a past affective environment study (Isaacowitz et al., 2015), we have not been able to separate choices from fixations because whereas stationary eye tracking records in two dimensions and permits nuanced analyses of fixations within dynamic stimuli (such as videos) presented on a screen, mobile eye tracking records in 3D space. In a mobile environment, when a participant looks at a stimulus presented on a screen, it is straightforward to process data showing that gaze is on the screen (attentional selection), but there is no existing process to consider nuanced fixations within the screen (attentional deployment). Put another way, in stationary eye tracking with existing programs, one can show a scary movie on the screen, and determine how much a participant fixated on the scariest clips of that video. In mobile eye tracking, existing programs (such as used in Isaacowitz et al., 2015) can determine that a participant has fixated on the general screen where scary movies are shown, but not how they fixated toward the scariest parts of the movies.

Before the development of the method described in this paper, analyzing specific fixations to particular components of dynamic stimuli freely chosen by a participant and recorded with a mobile eye tracker would have required manual annotation by a human coder watching the environment video for each participant, determining which fixations appeared in which LookZones in selected videos (after selecting a negative video, how much or little do they fixate on the most negative parts?). For example, a recent mobile eye tracking paper by Macdonald and Tatler (2015) used manual coding to determine when participants (N = 24) looked toward the instructor. However, this would not be feasible for an emotion regulation study, given the larger numbers of subjects and target stimuli. We therefore developed a novel approach.

Current Study: Assessing Situation Selection and Attentional Deployment Together

For the current project, we devised a method for recording and analyzing fixations to specific areas of selected stimuli within the mobile eye tracking environment, allowing us to assess both situation selection as well as attention deployment to selected stimuli. The basic premise of the approach, described in more detail below (see “Data Processing”), was to convert 3D coordinates recorded with mobile eye tracking into 2D coordinates of the selected stimulus that could then be analyzed with existing LookZone automation designed for stationary eye tracking.

Unlike studies of attentional selection, which analyzed fixation within the video more generally, this allows us to analyze fixations to LookZones in selected videos in a way that is parallel to how we analyze fixations to (fixed, experimenter-selected) videos in stationary tracking studies of attention. Thus, for the first time, we could test whether older adults select relatively more positive stimuli (situation selection) and whether they simultaneously show more positive fixation patterns towards those selected stimuli (attentional deployment). Do older people select negative content that is interesting, but then systematically avoid looking at the most negative parts of it? Because some past work has suggested that age differences (at least for attentional deployment) are most pronounced when participants are already in negative mood states (rather than simply their pre-experimental mood states: Isaacowitz et al., 2008), we induced negative mood states before the mobile eye tracking task.

Because SST posits that older adults’ positivity is due to motivational processes involving hedonic goals, we also instructed half the participants specifically to try to minimize their negative feelings while interacting with the choices. The other participants were instructed simply to view whatever was interesting to them. Thus, we could evaluate the effects of both age and motivation on both situation selection and attentional deployment. To this end, we measured choice behavior (situation selection) as well as eye fixation to emotionally salient material in selected choices (attentional deployment to selected situations). We also measured mood to determine the extent to which choices, and fixations to those choices, predicted mood. Given our interest in potential age differences in both the use and effectiveness of different emotion regulation strategies (see, for example, Isaacowitz & Blanchard-Fields, 2012; Livingstone & Isaacowitz, 2016) and in the role of motivation (Livingstone & Isaacowitz, 2015) we also considered whether either age or instructions moderated these effects.

Hypotheses

Certain predictions regarding whether attentional deployment to self-selected stimuli more like attentional deployment to experimenter chosen stimuli or more like situation selection can be formulated based on relevant conceptual models and past findings. If selection is paramount, the pattern should be like situation selection and yield few age differences. If attention is paramount, and older adults show robust positive attentional patterns regardless of whether stimuli are self-selected or experimenter-selected, there should be age differences with older adults showing more positive attentional patterns in our mobile eye tracking assessment of fixations. If positivity characterizes all older adults’ emotion regulation efforts, we would expect older adults to select more positive content and show a more positive gaze pattern (in other words, fixate less on the negative parts of negative selections, and more on the positive parts of positive selections) compared to younger and middle-aged adults. If positivity characterizes only older adults’ attentional deployment, we would expect no age difference in selections, but we would expect older adults to attend more to positive content and less to negative content within selections compared to other age groups.

In addition to considering selections and fixations to those selections, it is important to consider mood change as well to ascertain the effectiveness of these putative emotion regulatory behaviors (Isaacowitz & Blanchard-Fields, 2012). For either of the above possibilities, older adults’ moods would be expected to be best when they followed a more positive (selection/gaze) pattern. If motivation is more important than age, instructions to regulate should lead to more positive selections, fixation, and mood, regardless of age.

Method

Participants

Younger adults (YAs) were recruited from introductory psychology classes and given course credit for their participation. Middle-aged (MAs) and older adults (OAs) were recruited through online and printed advertisements; older adults were additionally recruited from a database of past volunteers. Community participants were paid $10/hour for their participation. We planned to recruit 150 participants (50 each of younger, middle-aged and older adults) based on a priori power analyses using previous stationary eye-tracking studies to estimate effect sizes.

Participants were 50 younger adults (ages 18–23, M_age = 19.30, SD = 1.63), 50 middle-aged adults (ages 35–59, M_age = 47.35, SD = 7.08), and 50 older adults (ages 60–86, M_age = 70.04, SD = 7.47). The sample was 41% male, 55% female (4% did not report); 70% White, 11% Black, 12% Asian, and 6% Hispanic. The study was approved by the Northeastern IRB.

Stimuli and Measures

The “Affective Environment” is a medium-sized room with three computer screens and a rolling chair that allows participants to move easily among them. Our interest was in attention once a particular stimulus is selected, rather than attention during the process of selecting one stimulus over another, so the set-up was optimized for that goal. The presence of three screens allows a menu of options from each valence (positive, negative, or emotionally neutral) to be visible simultaneously during selection and large enough to see from different locations. This set-up reduced the need for participants to search for options and maximized their ability to simply consider (and select) from the always-presented full set of alternatives. Each screen presented five videos of a different valence; position of valence was counterbalanced across subjects.

Video Stimuli

Stimuli were five positively valenced, five negatively valenced, and five emotionally neutral short video segments (0:50 to 3:43 minutes) downloaded from the internet and pre-rated for valence, arousal, and personal relevance by 10 YAs, 10 MAs, and 10 OAs. Pretesting rating data can be found in Supplemental Materials. Initially and between choices, a menu screen with five thumbnail images appeared on each screen, along with brief descriptions of the content (see Table 2). Participants could choose to view a video by clicking on the thumbnail with their mouse. Videos were presented using a program we created that recorded exact timing of start and stop times of videos and uploaded data to MongoDB servers (MongoDB Inc., New York, NY). Number of negative, neutral, and positive videos chosen served as the measure of situation selection.

Table 2.

Video Stimuli in the Affective Environment

Valence	Length	Description Seen by Participants
Positive	2:31	A second a day
Positive	3:43	Breathtaking earth scenes
Positive	0:50	Jonathan’s cochlear implant activation
Positive	2:43	Cutest bear attack ever
Positive	1:06	Dirty Dancing
Neutral	2:09	What would happen if the world lost its oxygen
Neutral	2:11	Cool science experiments you can do at home
Neutral	3:07	Basic happiness – extreme conditions
Neutral	3:39	How to prepare for a painting project – The Home Depot
Neutral	2:08	Incredible chemical reaction
Negative	1:26	Marley & Me
Negative	1:43	Polar bear attack
Negative	1:44	Top 10 animal fights caught on camera
Negative	2:06	Mother to 911- “I killed my kids”
Negative	3:33	The embalming process

Note. When in the affective environment, participants saw the captions above a thumbnail image depicting a representative scene within the video.

Mood Ratings

Before beginning the selection task and after viewing each video, participants rated their mood on the computer using 9-point valence and arousal scales depicting the Self-Assessment Manikin (SAM; Bradley & Lang, 1994) with verbal anchors. Higher numbers indicate more positive mood and higher arousal, respectively.

Eye Tracking

Eye tracking was used to assess attention to selected stimuli. Eye movements were recorded at a temporal resolution of 30 Hz, and an accuracy of 0.5° to 1.0° visual angle using an ASL MobileEye XG eye tracker (Applied Science Laboratories, Bedford, MA), which allowed participants to move among computers while wearing the eye camera mounted on a pair of glasses. Fixations were defined as holding a point of gaze for 100 ms within one visual degree. Areas of Interest (AOIs) were identified around the areas of each of the three computer screens to establish parameters from which to calculate fixations within LookZones (see below).

LookZones within Videos

A primary goal of this research was to assess attention—specifically positive gaze patterns (i.e., looking less at the most negative parts of a negative video, or more at the more positive parts of a positive video)—within videos that had been selected. LookZones were created within videos using GazeTracker software (EyeTellect LLC, Charlottesville, VA). Thirteen trained research assistants identified the most positive (for positive videos), negative (for negative videos), or visually salient (for neutral videos) areas of each video. If an area was mentioned by at least 33% of research assistants as falling into the target category (positive, negative, or salient), it was considered a LookZone. Between two and five LookZones were identified for each video. Negative (11% of the screen), neutral (14%), and positive (15%) videos did not differ in the size of LookZones, F(2,12) = .48, p = .629. Negative (16% of the video), neutral (16%), and positive (17%) videos also did not differ in the duration that they appeared, F(2,12) = .09, p = .913. The LookZones were thus comparable across video valences.

Procedure

After giving informed consent, participants completed a series of individual difference measures and a vision test. Participants were then fitted with the eye tracking glasses, and their gaze was calibrated across the three computers using a 12-point procedure. An experimenter provided instructions, led participants through a practice selection task and mood rating, and answered any questions.

Mood Induction

Before beginning the selection task, participants underwent a negative mood induction described in Lynn, Zhang, and Barrett (2012). Participants viewed a series of moderate- and high-arousal (M_arousal = 6.29, SD = .59) negative (M_valence = 2.18, SD = .45) images from the International Affective Picture System (IAPS, Lang, Bradley, & Cuthbert, 2005), accompanied by mood-congruent music, for 4.5 minutes.¹ Participants rated their mood before and after the mood induction.

Instruction Manipulation

Participants were randomly assigned to either a just view or regulate condition (n = 25 for each age group in each instruction condition). After the mood induction, participants in the just-view condition were told, “While you are in the room, your goal is to choose whatever is interesting to you.” Those in the regulate condition were told, “While you are in the room, your goal is to minimize your negative emotions or feelings.”

Selection Task

The researcher explained to participants that they would be alone in the room for 15 minutes, during which time they would be free to choose whichever of the stimuli on any of the three computers that they wanted to watch. They were not told the valence of the three computers. The experimenter left the room for 15 minutes while participants selected and viewed videos. Once chosen, the video played out in its entirety. After each video, participants rated their current valence and arousal. Participants continued to choose and view videos until the 15 minutes were up (M_choices = 5.31, SD = .97). Video choices, mood ratings, and eye data were recorded. After 15 minutes, the experimenter returned to the room and asked participants to recall the instructions they had been given before the task as a manipulation check. Participants were then compensated and debriefed.

Results

Data Processing

The mobile eye tracking system cannot assess fixations in LookZones within individual videos, which varied across participants due to the free-choice paradigm. To do this, we developed a database-oriented program to convert idiosyncratic 3D mobile eye data to output that could be analyzed in 2D. During data collection, we carefully synced time among the computers. Then we extracted information from the mobile eye tracking data about which computer screen participants were viewing and when, and the X and Y coordinates of fixations on each screen. Fixation times and coordinates were uploaded to MongoDB, a non-relational database program, where they could be integrated with data on the start and stop times of videos for each participant and compared to LookZone coordinates appearing within videos. A second program (LookZone Analysis) analyzed the data and output information on percentage of time participants fixated in LookZones when they were on the screen (thus recreating a typical stationary eye tracking data processing approach to video LookZones). For validation of this program, see Supplemental Materials.

Mood Induction Check

A 2 (time) x 3 (age group) ANOVA showed that participants reported more negative mood after the mood induction (M = 2.43, SD = 1.18) compared to before (M = 6.05, SD = 1.35), F(1,129) = 693.67, p < .001, η² = .843. There was also a main effect of age, F(2,129) = 8.28, p < .001, η² = .114; and an age x time interaction, F(2,129) = 4.84, p = .009, η² = .070. There was a significant age difference before the mood induction, F(2,129) = 11.90, p < .001, η² = .156: Younger adults (M = 5.29, SD = 1.00) reported more negative mood compared to both middle-aged (M = 6.51, SD = 1.39) and older adults (M = 6.31, SD = 1.31), ps < .001; the latter two groups did not differ (p = .449). After mood induction, there were no differences among younger (M = 2.29, SD =.94), middle-aged (M = 2.64, SD = 1.15), and older (M = 2.36, SD = 1.38) adults’ mood, F(2,129) = 1.15, p = .320, η² = .018.

Positivity in Situation Selection

We first tested whether there was an age-related positivity effect in situation selection, using a 3 (age group) x 2 (instruction condition) x 3 (valence of selection) ANOVA. These analyses examined whether any individual differences in situation selection could be attributed to age, motivation, or both. The dependent variable was the number of videos that participants viewed in each valence category. Descriptive statistics for all participants are reported in Table 3.

Table 3.

Mean number of negative, neutral, and positive videos chosen by age and instruction condition

	Negative	Neutral	Positive
YA Just View	1.28 (1.31)	2.28 (1.17)	2.20 (1.15)
YA Regulate	0.88 (1.24)	2.04 (0.98)	2.44 (1.39)
YA Total	1.08 (1.28)	2.16 (1.08)	2.32 (1.15)
MA Just View	1.64 (1.25)	1.84 (1.11)	2.08 (1.41)
MA Regulate	1.24 (1.20)	1.84 (1.03)	2.20 (1.12)
MA Total	1.44 (1.23)	1.84 (1.06)	2.14 (1.26)
OA Just View	0.88 (1.09)	1.76 (1.33)	2.48 (1.23)
OA Regulate	0.84 (0.90)	1.88 (1.20)	2.04 (1.51)
OA Total	0.86 (0.99)	1.82 (1.26)	2.26 (1.38)
Just View Total	1.27 (1.24)	1.96 (1.21)	2.25 (1.26)
Regulate Total	0.99 (1.12)	1.92 (1.06)	2.23 (1.34)
Total	1.13 (1.19)	1.94 (1.14)	2.24 (1.30)

Note. Standard deviations are presented in parentheses. YA = younger adult, MA = middle-aged adult, OA = older adult.

The main effect of valence was significant, F(2,288) = 24.23, p < .001, η² = .144, with participants choosing fewer negative videos than neutral and positive, which did not significantly differ. There were no significant interactions between valence and age group, F(4,288) = 1.16, p = .329, η² = .016; instruction condition, F(2,288) = .37, p = .690, η² = .003; or age x instruction, F(4,288) = .63, p = .641, η² = .009.

There was a significant main effect of age, F(2,144) = 6.10, p = .003, η² = .078, and a significant main effect of instruction condition, F(1,144) = 5.20, p = .024, η² = .035. Older participants made fewer choices (M = 4.94, SD = .98) than younger (M = 5.56, SD = 1.01) and middle-aged participants (M = 5.42, SD = .81), who did not differ. Participants in the just view condition made more choices (M = 5.48, SD = 1.07) than those in the regulate condition (M = 5.13, SD = .83). When examining percent of choices (which controls for group differences in number of choices), the pattern was the same.

Positivity in Attentional Deployment toward Self-Selected Stimuli

We then tested whether, within the selections participants made, there was an age-related positivity effect in attention to LookZones—the most positive, negative, or visually salient parts of the images within the selections they made. Because not all participants chose videos from all three valence categories, we tested this question using multilevel modeling in HLM 7 (Raudenbush, Bryk, & Congdon, 2013), with video selections (Level 1) nested within participants (Level 2). Percent fixation within LookZones was modeled as a function of video valence (Level 1), age group and instruction condition (Level 2), and their interactions. Age, instruction, and valence were dummy coded so that younger adults, the just view group, and neutral valence were the reference categories (coded as zero).² Random effects were included in the initial model but were removed if they were not significant at p < .10 (Nezlek, 2011).³

The final model was as follows:

$$\begin{gathered} \text{Fixation}\text{in}\text{LookZones}={\beta }_0+{\beta }_1(\text{NEG})+{\beta }_2(\text{POS})+r \\ {\beta }_0={\gamma }_{00}+{\gamma }_{01}(\text{INSTR})+{\gamma }_{02}(\text{MA})+{\gamma }_{03}(\text{OA})+{\gamma }_{04}(\text{MA}\ast \text{INSTR})+{\gamma }_{05}(\text{OA}\ast \text{INSTR})+u_0 \\ {\beta }_1={\gamma }_{10}+{\gamma }_{11}(\text{INSTR})+{\gamma }_{12}(\text{MA})+{\gamma }_{13}(\text{OA})+{\gamma }_{14}(\text{MA}\ast \text{INSTR})+{\gamma }_{15}(\text{OA}\ast \text{INSTR}) \\ {\beta }_2={\gamma }_{20}+{\gamma }_{21}(\text{INSTR})+{\gamma }_{22}(\text{MA})+{\gamma }_{23}(\text{OA})+{\gamma }_{24}(\text{MA}\ast \text{INSTR})+{\gamma }_{25}(\text{OA}\ast \text{INSTR}) \end{gathered}$$

Participants were included in analyses if their gaze was tracked (i.e., the eye tracking system could determine the position and orientation of the participant’s pupil) for at least 75% of the time. Data were available for 648 videos (Level 1) across 130 participants (Level 2). On average, there were 4.98 (SD = 1.09) observations per person (range 2 – 7), with 21.13% of variance in gaze between-subjects. The full model explained 3.5% of the variance in fixation in LookZones (R₁²; Snijders & Bosker, 2011).

Results are reported in Table 4. There were no main effects of age or instruction condition, and no age x instruction interaction (ps > .25), indicating that overall fixation in LookZones did not differ as a function of age or instruction condition. Overall, people fixated less in negative LookZones than in neutral, t(506) = −2.30, p = .022, but neutral and positive did not differ, t(506) = −.94, p = .347. There was also a significant MA x positive valence interaction, which we note was not hypothesized and must be interpreted with caution. As an exploratory analysis, we conducted analyses separately for neutral and positive valence. For neutral choices, there were no significant group differences or interactions, ts < 1.5, ps > .20. Middle-aged adults, however, spent more time fixating in LookZones in positive videos than younger adults, t(109) = 3.87, p < .001. No other effects were significant, ts < 1.5, ps > .20.

Table 4.

Results from Multilevel Model Predicting Fixation in LookZones from Age Group, Instruction Condition, and Video Valence

Fixed Effect	Coefficient	SE	t-ratio	df	p-value
Intercept γ00	33.24	3.36	9.89	124	<0.001
Instruction condition γ01	−2.78	4.71	−0.59	124	.557
MA γ02	−3.92	5.36	−0.73	124	.446
OA γ03	−5.45	5.41	−1.01	124	.316
MA x Instruction γ04	8.48	7.44	1.14	124	.256
OA x Instruction γ05	1.57	7.35	0.21	124	.831
Negative valence γ10	−11.25	4.89	−2.30	506	.022
Negative x Instruction γ11	13.86	7.97	1.74	506	.083
Negative x MA γ12	11.01	6.43	1.71	506	.088
Negative x OA γ13	2.04	6.33	0.32	506	.747
Negative x MA x Instruction γ14	−13.25	10.40	−1.27	506	.203
Negative x OA x Instruction γ15	−7.16	9.92	−0.72	506	.471
Positive valence γ20	−4.29	4.55	−9.94	506	.347
Positive x Instruction γ21	2.85	5.52	0.52	506	.606
Positive x MA γ22	16.80	5.53	3.04	506	.002
Positive x OA γ23	5.31	6.24	0.85	506	.395
Positive x MA x Instruction γ24	−15.04	7.82	−1.93	506	.055
Positive x OA x Instruction γ25	−3.57	8.07	−0.44	506	.658

Note. YA = younger adult, MA = middle-aged adult, OA = older adult.

Effects of Situation Selection on Mood

To examine the relationships between age, instruction condition, situation selection, and mood, we ran a series of hierarchical linear regressions. Rather than provide choice-level information about how individual selections are related to mood, these analyses characterize the entire session, testing the more global relationship between selections across the 15-minutes and mood. Mood, averaged across the situation selection task, served as the dependent variable. Dummy codes were created for middle-aged and older adult groups, as well as for the regulate instruction condition. Baseline mood (post-mood induction) was not a significant predictor of task mood in any of the models tested below (as a single predictor, β = .076, p = .385), so it was not included in analyses.

Between-group dummy codes (age and instruction condition) and their interactions were entered first. Number of negative and positive choices (centered) were entered second. Interactions between choices and individual difference dummy codes were entered third. Three-way interactions were entered fourth.

Complete results can be found in Table 5. In Step 1, older adults reported marginally more positive mood over the session compared to younger adults (β = .258, p = .054); this effect became weaker (ps > .07) when selections were included. There was no effect of instruction condition in any of the models (βs = .084 to .188, ps > .15). There was a robust relationship between number of negative choices and mood (βs = −.446 to −.585, ps < .01). The relationship between number of positive choices and mood was significant in Step 2 (β = .213, p = .003), but not when interactions were included (ps = .067, .145).

Table 5.

Hierarchical Multiple Regression Analyses Predicting Mood from Age, Instruction, and Choices

Predictor	Step 1 β	Step 2 β	Step 3 β	Step 4 β	Model F	Model R2	R2 change
Instruction	.188	.085	.090	.084
MA	.008	.088	.148	.194
OA	.258	.158	.165	.187
MA x Instruction	−.057	−.049	−.107	−.139
OA x Instruction	−.170	−.072	−.035	−.043	1.34	.044	.044
Negative Choices		−.494***	−.585***	−.456**
Positive Choices		.213**	.271	.264	13.29***	.396	.351**
MA x Negative			−.180	−.361**
OA x Negative			.080	.040
MA x Positive			−.151	−.119
OA x Positive			−.078	−.142
Instr. x Negative			.272**	.023
Instr. x Positive			.153	.058	8.58***	.451	.055*
MA x Instr. x Neg				.292*
OA x Instr x Neg.				.085
MA x Instr. x Pos.				−.010
OA x Instr. x Pos.				.145	7.13***	.479	.028

Note. YA = younger adult, MA = middle-aged adult, OA = older adult; Instr = Instruction condition; Neg = number of negative choices; Pos = number of positive choices.

^*p < .05,

^**p < .01,

^***p < .001

There was a significant interaction between instruction condition and negative choices in Step 3 (β = .272, p = .007), which was not significant when including three-way interactions. Follow-up correlations suggested that the relationship between number of negative choices and mood was stronger for those in the just view condition, r(73) = −.66 than for those in the regulation condition r(73) = −.49, though both were significant, ps < .001, and these correlations did not differ significantly, z = −1.45, p = .147.

In Step 4, there were significant interactions between middle-age and negative choices, and between middle-age, negative choices, and instruction condition. Follow-up correlations showed that the relationship between number of negative choices and mood was significant and strong for middle-aged adults in the just view condition, r(23) = −.83, p < .001, but not for those in the regulate condition, r(23) = −.34, p = .092, the correlations were significantly different, z = −2.77, p = .006.

Effects of Attentional Deployment to Self-Selected Stimuli on Mood

We then tested whether fixation was related to mood and whether this varied by age or instruction condition, using multilevel modeling. Mood data were available for 796 observations across 150 participants (648 observations also had eye tracking data), with 19.38% of the variance in mood between-subjects. The full model explained 8.3% of the variance in video-level mood (R₁²; Snijders & Bosker, 2011).

As before, dummy codes were created for age group, instruction condition, and video valence (younger adults, the just view group, and neutral videos = 0). Fixation was entered grand-mean centered.⁴ As in the regression analysis, baseline mood (post-mood induction, grand-mean centered) was not a significant predictor of mood ratings during the 15 minutes (γ ₀₁ = .067, SE = .100), t(130) = .77, p = .503, and so was not included in these analyses. Random effects were initially modeled but removed if variance components were not significant at p < .10 (Nezlek, 2011). Only the intercept and the slope of negative mood were modeled as randomly varying; the other Level 1 effects were modeled as non-randomly varying (as a function of Level 2 predictors; see Nezlek, 2011). Several of the random effects were not significant, suggesting that variance in slopes cannot be reliably estimated based on current information. We therefore dropped the random effects from the model (deleting the smallest effect one at a time), and removed relevant cross-level interaction terms from the model.⁵ The final model was as follows:

$$\begin{gathered} \text{Mood}={\beta }_0+{\beta }_1(\text{NEG})+{\beta }_2(\text{POS})+{\beta }_3(\text{FixationInLZs})+{\beta }_4(\text{FixXNeg})+{\beta }_5(\text{FixXPos})+r \\ {\beta }_0={\gamma }_{00}+{\gamma }_{01}(\text{INSTR})+{\gamma }_{02}(\text{MA})+{\gamma }_{03}(\text{OA})+{\gamma }_{04}(\text{MA}\ast \text{INSTR})+{\gamma }_{05}(\text{OA}\ast \text{INSTR})+u_0 \\ {\beta }_1={\gamma }_{10}+{\gamma }_{11}(\text{INSTR})+{\gamma }_{12}(\text{MA})+{\gamma }_{13}(\text{OA})+{\gamma }_{14}(\text{MA}\ast \text{INSTR})+{\gamma }_{15}(\text{OA}\ast \text{INSTR})+u_1 \\ {\beta }_2={\gamma }_{20} \\ {\beta }_3={\gamma }_{30} \\ {\beta }_4={\gamma }_{40} \\ {\beta }_5={\gamma }_{50} \end{gathered}$$

Results are presented in Table 6. There were no main effects of age or instruction condition, and no age x instruction interaction, ps > .25. Across all participants, mood was more negative after negative videos compared to neutral, t(124) = −6.67, p < .001; and more positive after positive compared to neutral, t(384) = 10.24, p < .001. The relationship between fixation and affect was not significant, t(384) = 0.26, p = .793, nor were the interactions between fixation and negative valence, t(384) = −0.86, p = .391; and positive valence, t(384) = 0.72, p = .475. Thus, the hypothesis that the relationship between fixation and affect would vary by age and instructions was not supported.

Table 6.

Results from Multilevel Model Predicting Affect from Age Group, Instruction Condition, Video Valence, and Fixation in LookZones

Fixed Effect	Coefficient	SE	t-ratio	df	p-value
Intercept γ00	6.01	0.18	33.45	124	<.001
Instruction condition γ01	0.26	0.23	1.15	124	.251
MA γ02	0.05	0.27	0.20	124	.844
OA γ03	0.29	0.26	1.10	124	.275
MA x Instruction, γ04	−0.13	0.38	−0.34	124	.733
OA X Instruction, γ05	−0.24	0.35	−0.69	124	.495
Negative Valence γ10	−2.10	0.31	−6.67	124	<.001
Negative x Instr. γ11	0.18	0.70	0.26	124	.798
Negative x MA γ12	−0.55	0.44	−1.23	124	.222
Negative x OA γ13	−0.56	0.52	−1.07	124	.286
Neg x MA x Instr γ14	0.15	0.90	0.16	124	.872
Neg x OA x Instr. γ15	0.66	1.02	0.65	124	.515
Positive Valence γ20	1.05	0.10	10.24	384	<.001
Fixation γ30	0.001	0.004	0.26	384	.793
Negative x Fixation γ40	−0.007	0.008	−0.86	384	.391
Positive x Fixation γ50	0.004	0.005	0.72	384	.475

Note. Higher-order effects are in bold. YA = younger adult, MA = middle-aged adult, OA = older adult.

Discussion

Lab studies of emotion regulation tend to study the use and effectiveness of emotion regulation strategies without considering how use of that strategy might be constrained by preferences for other strategies. In the context of aging, studies of attentional deployment have provided consistent evidence for age differences, with older adults looking relatively less at negative and more at positive stimuli and these gaze patterns sometimes predicting better mood (see Isaacowitz, 2012 for a review). In contrast, studies of situation selection have found more age-similarity in choices and their influence on mood (e.g., Rovenpor et al., 2013; Livingstone & Isaacowitz, 2015). In everyday life, attentional deployment is not independent of situation selection; what we attend to is constrained by what we have chosen to expose ourselves to. In the current research, we used mobile eye tracking and a flexible database system to analyze fixation data to consider potential age differences in the dynamic interplay of situation selection (video choices) and attentional deployment (fixation to emotionally salient areas) when regulating out of negative mood states. This allowed us to determine whether attention to self-selected stimuli is more like attention to experimenter-selected stimuli (age differences in positivity) or more like selection-based strategies (age similarity in positivity).

Age Similarities in Attentional Deployment to Self-Selected Stimuli

Overall, our findings from attention to self-selected stimuli are more similar to the selection-based findings (Sands et al., in press) than the attention to experimenter-selected stimuli findings (Isaacowitz, 2012); see Table 7 for a summary. Age differences consistently emerge when emotional stimuli are presented by experimenters. In these cases, the earliest stage in which to influence one’s emotions according to the process model is attentional deployment: Participants can look more or less at them, or more or less at the most emotionally evocative parts. In these studies, older adults consistently look less at negative content (and sometimes more at positive; see Isaacowitz, 2012). In contrast, the current research shows that allowing participants to select the content they can attend to (or not) leads to fewer age differences.

Table 7.

Summary of Current Findings Regarding Situation Selection and Attentional Deployment

Construct	Operationalization	Current Findings: Use	Current Findings: Relationship to Mood
Situation Selection	Number of negative, neutral, and positive videos chosen	People chose fewer negative videos than neutral or positive	Robust relationship between negative choices and mood; this was stronger for those in the just view condition. The relationship between positive choices and mood was present but weaker.
Attentional Deployment	Percent fixation to emotional (or informationally salient for neutral) LookZones within chosen video	People fixated less in negative LookZones than in neutral and positive [Middle-aged adults fixated more in positive LookZones than younger adults.]	Fixation was not related to mood when controlling for video valence, some interactions with other variables [No relationship in positive videos; weak relationships in negative videos; relationship varied by age in neutral: for the JV group, greater fixation was related to more positive affect, whereas the relationship was not significant for MAs and OAs in the REG group, and related to less positive affect for YAs in the regulate group.]

Note. Results in brackets should be considered exploratory analyses. Attentional selection was assessed in this project but was not a major variable of interest given our focus on actual attentional deployment to selected stimuli; therefore, we have not included it in the results or in this summary, but for comparison with our review of past findings, the comparable attentional selection findings can be found in Supplemental Materials.

This pattern suggests there is less need for older adults to use attentional deployment as their signature emotion regulatory strategy when potential elicitors in the environment are freely chosen and thus intrinsically interesting to older adult perceivers. We did not, for example, find evidence that older adults showed a looking pattern suggesting they were “looking through their fingers” or “peeking” at interesting but potentially disturbing stimuli. In exploratory analyses, however, we did find that middle-aged adults tended to focus on the most emotionally salient parts of positive videos they had chosen, compared to younger and older adults. Together with previous findings that middle-aged adults may show even greater positivity in attention compared to older adults in some contexts (Isaacowitz & Harris, 2014), this suggests that middle-aged individuals may sometimes show the largest age-related positivity effects in attention, perhaps due to greater attentional resources or stronger context-specific hedonic goals (or a combination of the two). Interestingly, middle-aged individuals also seemed to shift behavior in line with instructions; in particular, they seem to shift away from using selection-based strategies when instructed to regulate. With so little research on middle-aged adults’ emotion regulation, future work is needed to better understand the emotion regulation strategy use of this age group specifically.

We did not find consistent age differences in situation selection. This is consistent with some past studies on age differences in situation selection (Isaacowitz et al., 2015; Rovenpor et al., 2013; Sands et al., in press). In the current study, we manipulated mood before starting the task, because age-differences in attentional deployment are more apparent when participants are already in negative moods they might like to get out of (see Isaacowitz et al., 2008). This may not be the case for situation selection, however. We found that participants of all ages preferred positive and neutral over negative stimuli when in a negative mood, suggesting some basic situation selection behavior when in an induced negative mood.

The Role of Motivation

We also manipulated instructions in order to disentangle the effects of age and motivation on emotion regulation (see Livingstone & Isaacowitz, 2015), due to past findings that age differences may be more likely to emerge when participants are specifically told to regulate how they feel (vs. to look naturally, e.g., Isaacowitz & Choi, 2012). If motivation is a driving factor in the use of emotion regulation, we would expect similar results across age groups for people within the regulate condition and more variability in age effects on emotion regulation in the just-view condition. Some effects of instructions on emotion regulation did emerge: For example, compared to those told to pick what was interesting to them, participants asked to regulate made fewer selections. In general, though, regulation instructions did not influence fixation or mood for younger and older adults. It appears that the effects of the negative mood induction may have motivated participants to regulate, thus outweighing potential effects of the regulation instructions.

In their meta-analysis, Reed, Chan, and Mikels (2014) argued that age-related positivity effects were most prominent in relatively less constrained tasks, as greater task instructions would give older adults less opportunity to display their inherent positivity (see also Reed & Carstensen, 2012). Our results suggest this may not apply to situation selection: When participants choose the nature of their emotional input themselves—a relatively unconstrained task—participants in all age groups preferred positivity and did not show the typical age-related positivity in attentional deployment typically found in fixed-stimulus studies.

Effectiveness of Situation Selection and Attentional Deployment

Across all age groups in both instruction conditions, negative choices were consistently related to more negative mood. Positive selections had a strong effect at the video level, but a weaker one across the 15-minute session. The relationship between fixation and mood was less clear. Overall, fixation was not significantly related to mood. In addition, because there was little variance in the relationship between fixation and mood, we were not able to test whether this relationship varied across age and instruction groups.

In sum, situation selection appears to have had a stronger and more consistent effect on mood than attentional deployment; the latter was not related in the consistent way that studies of stationary eye tracking using fixed stimuli have found. Mobile and stationary eye tracking, therefore, yield different age-related patterns.

Limitations and Future Directions

One limitation of the current study that future work should address is the potential role of individual difference variables. We did not have adequate power in this sample to consider individual differences, given that we were already considering age and instruction condition as between-subject factors. However, past work suggests a variety of individual differences that may be relevant. For example, work on attentional deployment has sometimes found general attentional abilities to be a moderator of fixation-moods links (e.g., Noh et al., 2011) and has found divided attention to influence the nature of age-related positivity effects in attention (Knight et al., 2007). One study found that beliefs about emotion regulation moderated certain age differences in situation selection behavior (Rovenpor et al., 2013). Importantly, the effects of these individual differences may further vary as a function of age. Future studies will need quite large sample sizes to consider various cognitive, affective, and personality individual differences that could be relevant.

Another limitation involves the finite set of stimuli that we used as choices. Though we did extensive pretesting to ensure that this set contained options that were comparable across age groups, future work could use different stimulus sets to make sure the findings generalize to other choice stimuli. We also focused on attention once a stimulus was selected rather than attention during the process of making a selection; that would require a different design but could be of interest for future work.

A final limitation is that we only used a negative mood induction in the current study, in order to maximize the need for regulation in our participants, and also given past findings that age differences in attention are greatest during initial negative mood states (Isaacowitz et al., 2008). However, we have also run an identical study with a positive rather than negative mood induction and found a similar pattern of minimal age differences (see Supplemental Materials), suggesting our findings are not an artifact of the negative induction. It remains an open question, however, what pattern we might find if participants did not experience a mood induction before completing the task.

Conclusions: Implications for Studying Emotion Regulation across Adulthood

These findings lead to the tentative conclusion that studies of single-stream attentional processes may overstate the magnitude of age-related positivity effects in real life. Such studies reflect a context in which older adults may be exposed to emotional stimuli that are of little interest to them, as they have not had a role in choosing to engage with them. Thus, age-related positivity effects in attention may be most often found, and may play a larger role in, emotion regulation in those situations (mimicked in the lab) where perceivers cannot control what they are exposed to. In everyday life, where perceivers have more discretion regarding the contents of their environment, younger and older adults show similar choice and attention patterns with similar overall effects on real-time mood regulation. Lab studies of emotion regulation that do not permit perceivers to choose their environment may therefore serve to amplify potential group differences.

These findings have further implications for studying the process model of emotion regulation. The original model proposed that situation selection is the earliest stage at which a person can intervene in the emotional process (Gross, 1998). This study provides evidence that all age groups do engage in situation selection when motivated (see also Livingstone & Isaacowitz, 2015): There was a general tendency to avoid negative stimuli. The use of subsequent emotion regulation strategies (attentional deployment, cognitive reappraisal) depends on the situations that a person finds themselves in (see also the SOC-ER model: Urry & Gross, 2012), yet research on situation selection has lagged behind research on other forms (Webb, Miles, & Sheeran, 2012). Though our lab has begun to explore situation selection in the lab, more systematic research is needed to investigate the frequency of situation selection in everyday life. Work finding that the initiation of reappraisal is not very frequent (Suri, Whittaker, & Gross, 2015) further supports the need to consider the potentially more frequent use of selection as a key emotion regulation strategy at any age.

The choices a person makes have implications for the use of other forms of emotion regulation within the situations that are chosen; situation selection can also make subsequent emotion regulation unnecessary (see also Duckworth, Gendler, & Gross, 2016). Though we found no evidence of negative avoidance in looking once stimuli had been chosen, there was some evidence for middle-aged adults paying particular attention to the most emotionally relevant parts of positive stimuli, compared to younger adults. Further research is needed to better understand middle-aged adults and their emotion regulation strategy use in particular (e.g., Isaacowitz & Harris, 2014). Tools like mobile eye tracking have permitted us to consider how the use of some emotion regulation strategies constrains the use and effectiveness of others, and how these dynamic real-time processes may vary by age. Our findings from mobile eye tracking suggest subtle rather than pervasive age differences when individuals can select their own emotional environments.

A final technical implication is that our solution to the limitations of mobile eye tracking for assessing fixations to self-selected stimuli was successful but ultimately did not yield substantially different findings from previous work on attentional selection, which uses a much more crude measure from mobile tracking. Though our technical solution helped us ask the conceptual questions of interest for this study, the results suggest that the technical solution is not necessary for future work: Instead, the cruder mobile tracking measure or even the much simpler behavioral choice measure may be adequate for future work on aging and self-selected affective environments.

Figure 1.

Percent Fixation in LookZones as a function of age group, instruction condition, and video valence. JV = Just View condition, Reg = Regulate condition.