FABBS Doctoral Dissertation Research Excellence Award

Colleen Frank, University of Michigan, Department of Psychology

“The Role of Working Memory for Emotion in Affective Forecasting”

Colleen Frank, University of Michigan, Department of Psychology

Abstract

A crucial aspect of making good decisions is predicting how different outcomes will make us feel. From deciding which job to take, which house to buy, or who to marry, the ability to predict our future feelings is fundamental to wellbeing. This ability, known as affective forecasting, varies greatly among individuals, however, the reasons for this variability are poorly understood. This dissertation aims to uncover why some people are better forecasters than others. We test the hypothesis that affective forecasting depends specifically on affective working memory, a distinct domain of working memory responsible for actively maintaining and working with feeling states. If this hypothesis is correct, then individual differences in the ability to maintain emotional experiences should predict variability in the accuracy of affective forecasts. Across a series of four studies, we find that the ability to maintain emotions reliably predicts forecasting accuracy, whereas the ability to maintain perceptual (non-affective) information, does not. This selective association was further replicated using a real-life forecasting task based on the 2020 United States presidential election. These results demonstrate that affective working memory is a core mental ability that supports affective forecasting. Interventions that target affective working memory may, in turn, improve forecasting accuracy.

Introduction

Choice is a fundamental part of our daily lives—Should I ride my bike to work today? Should I take that new job across the country? Should I splurge on that fancy television? When making decisions, we often consider the anticipated emotions associated with each outcome, in order to select which option feels right (Mellers, Schwartz & Ritov, 1999; Charpentier et al., 2016). This ability to predict our future feelings is known as affective forecasting (Wilson & Gilbert, 2003). While errors (i.e., discrepancy between predicted and experienced feelings) in these predictions often lead to suboptimal decisions, there is a great deal of individual variability in forecasting accuracy (Dunn et al., 2007; Hoerger et al., 2012; 2016). Thus, an important question arises—what makes someone an accurate predictor of their future feelings? While previous studies have proposed that forecasting accuracy may rely on individual differences in trait-level abilities (i.e., emotional intelligence, personality traits), we propose that affective forecasting is also supported by a more fundamental psychological ability—i.e., affective working memory. The current work investigates affective working memory as a potential mechanism underlying affective forecasting. Affective working memory is a distinct domain of working memory responsible for actively holding emotion representations in mind (Davidson & Irwin, 1999; Mikels et al., 2005; Mikels & Reuter-Lorenz, 2019). Because affective forecasting requires conjuring up and maintaining emotional experiences for evaluation, we hypothesized that individuals who are better able to maintain feeling states (i.e., better affective working memory), would be more accurate in predicting their future feelings. Across two initial experiments (Studies 1 & 2) in the dissertation, results emerged in support of this hypothesis such that working memory for emotion uniquely predicted affective forecasting accuracy, whereas working memory for perceptual (i.e., brightness) information, did not (Frank et al., 2020). I focus here on two further pre-registered studies (Studies 3 & 4), that aim to further characterize the role of affective working memory in affective forecasting abilities. In Study 3, we more firmly establish the selective relationship between forecasting accuracy and affective working memory— a relationship not evident with cognitive working memory. We do so by testing whether performance on two additional, widely used measures of visual working memory explains any variability in forecasting accuracy beyond the variability explained by affective working memory (Frank et al., 2020). Study 4 examines whether the relationship between affective working memory and forecasting accuracy generalizes to a real-world event by measuring predicted and experienced feelings to the outcome of the 2020 United States presidential election. Moreover, unlike the prior studies in this dissertation which were all conducted in-person, with college students run in a laboratory setting, Study 4 was conducted entirely on-line with participants recruited nationwide, thereby establishing another dimension to the generality of the effects. To foreshadow our results, we find that affective working memory is unique in its ability to predict individual differences in affective forecasting accuracy, and that this relationship extends to forecasting tested using a real-world event.

The protocol for Study 3 and Study 4 can be found in Figure 1.

Figure 1. Protocol for Study 3 (top row) and Study 4 (bottom row). In Study 3, participants completed the affect maintenance, brightness maintenance, laboratory-based forecasting, and two cognitive working memory tasks across two sessions. In Study 4, a new sample of participants completed the affect maintenance, brightness maintenance, laboratory-based (administered on-line) forecasting and event-based forecasting tasks across three total sessions.

Affect Maintenance. Affective working memory is assessed with an affect maintenance task (Figure 2; Mikels et al., 2005; 2008; Broome et al., 2012). For each trial, participants view one emotional image (5 seconds). After it disappears, participants continue to actively hold the feelings elicited from the photo, at the same intensity level (3 seconds). They then view another emotional image (5 seconds), before deciding if the second image evoked more or less ‘emotional intensity’ (i.e., strength or amount of emotional reaction), compared to the first. To calculate maintenance task accuracy, participants subsequently rate the emotional intensity of each image separately. These ratings are used to infer which image, when appearing with its pair during the maintenance task, should be judged as evoking a more intense feeling.Brightness Maintenance. Non-affective maintenance abilities are assessed with a brightness maintenance task (Figure 1), designed as an emotionally-neutral analogue task requiring intensity maintenance (Mikels et al., 2005; 2008; Broome et al., 2012). For each trial, participants view one neutral image (5 seconds). After it disappears, participants continue to actively hold the brightness intensity in mind, at the same level (3 seconds). They then view a second neutral image (5 seconds), before deciding if the second image evoked more or less brightness intensity (i.e., amount of overall light or illumination), compared to the first. Similar to the affect maintenance task, participants rate the brightness of each image separately and these ratings are used to calculate maintenance accuracy.

Figure 2. Schematic for the maintenance tasks used. Participants hold the emotional or brightness intensity of one image in mind over a delay, before determining if a second image has more or less intensity.

Laboratory-Based Forecasting. To measure affective forecasting ability in the laboratory, participants read descriptions of emotional scenes and predict how they would feel if they were to view each image using a visual analog scale ranging from “extremely unpleasant” to “extremely pleasant” (Phase I; Hoerger et al., 2012). One-week later, participants return to view images of the emotional scenes and rate the feelings they experience using the same scale (Phase II). We then calculate the absolute difference between the predicted and experienced ratings, averaged across trials, to obtain a forecasting error score for each participant.

Cognitive Working Memory (Study 3). Non-affective (i.e., cognitive) working memory was assessed by two additional widely used measures of visuospatial working memory: visual change detection (Luck & Vogel, 1997) and Corsi block-tapping (Corsi, 1972).

Event-Based Forecasting (Study 4). To measure affective forecasting that better captures real-life affective predictions, we adapted an event-based forecasting measure to compare predicted and experienced feelings to the outcome of the 2020 United States presidential election (Lench et al., 2019). Participants first predicted how they would feel about the two major-party candidates winning the election (Phase I). After the election was called in favor of Joe Biden, participants rated their experienced feelings using the same scale (Phase II). Forecasting error was calculated as the absolute difference between ratings during Phase I and Phase II for each participant.

Results

A.

B.

Figure 3. Results from Study 4. A. Laboratory-based forecasting error was significantly predicted by affect maintenance (left), but not brightness maintenance (right), accuracy. This replicates the findings from Study 3 (not-pictured). B. Event-based forecasting error was significantly predicted by affect (left), but not brightness (right), maintenance accuracy.

Study 3 aimed to replicate and extend the initial findings of a selective association between affect maintenance and forecasting accuracy by assessing the potential role of cognitive working memory more broadly. The multiple linear regression included (laboratory-based) forecasting error as the outcome variable, and performance on the affect maintenance, brightness maintenance, and two additional cognitive working memory measures (visual change detection, Corsi block-tapping) as predictors. Consistent with our previous results, affect maintenance accuracy significantly predicted forecasting error (β = –.25, p = .022), whereas brightness maintenance accuracy did not (β = –.03, p = .802; Frank et al., 2020). Furthermore, while visual change detection performance predicted forecasting error in the regression model (β = .22, p = .049), this relationship no longer held when tested separately, r(82) = .19, p = .088—indicating that this is unlikely to be a true association. Additionally, Corsi block-tapping performance did not predict forecasting error, β = .13, p = .23 (Frank et al., 2020).

Study 4 examined whether affect maintenance, but not brightness maintenance, ability predicted people’s accuracy at forecasting their feelings about the outcome of the 2020 presidential election (event-based forecasting), in addition to the laboratory-based forecasting measure used all along. A multivariate linear regression was used with laboratory-based and event-based forecasting errors as the outcome variables and affect maintenance and brightness maintenance performance as the predictors. Critically, affect maintenance significantly predicted forecasting error on both tasks (laboratory-based; β = –.292, p = .023; event-based: β = –.33, p = .009; see Figure 3), such that higher accuracy on the affect maintenance task predicted more accurate forecasts. In contrast, brightness maintenance performance failed to predict forecasting error on either task (laboratory-based: β = –.084, p = .503; event-based: β = .177, p = .149).

Discussion

We reasoned that because affective forecasting requires conjuring up and maintaining emotional experiences for evaluation, affective working memory—the capacity to actively hold feeling states in mind—would play a key role in this ability. Supporting this hypothesis, the results from two pre-registered studies demonstrate that individual differences in affect, but not brightness, maintenance performance predict forecasting accuracy. The specific role of affective working memory is further documented by including additional measures of visuospatial working memory, which do not reliably predict forecasting accuracy (Study 3). Finally, this significant association holds true even when forecasting feelings about a major real-world event (Study 4). These results have implications for understanding the construct of affective working memory and the nature of affective forecasting.

The selective relationship among maintenance abilities and forecasting is especially informative, in that affective working memory appears to have a unique role in affective prediction accuracy. Across all studies, visual working memory did not predict forecasting accuracy when measured by the brightness maintenance, nor the Corsi block-tapping, tasks (Study 3). While better performance on the visual change detection task was associated with more forecasting errors in the regression model, this association did not remain significant in an independent correlation and thus, is unlikely to reflect a true association (Study 3). The lack of a relationship between cognitive working memory and forecasting accuracy underscores that the ability to actively maintain emotional experiences may be a core component of the ability to predict future feelings.

Since its inception, researchers have theorized that affective working memory may be related to a variety of higher-order emotion-related psychological processes including emotion regulation, wisdom, and even rumination (Mikels & Reuter-Lorenz, 2013; 2019). The present evidence reveals a direct association between affective working memory and an important form of prospective thought. This finding supports the idea that affective working memory is a fundamental psychological process and may support other forms of higher-order emotional thought. According to this formulation, just as cognitive working memory provides the mental workspace for complex cognitive tasks, affective working memory provides the mental workspace for mental simulations where feelings play a prominent role.

This work also has important implications for affective forecasting. In general, inaccurate prospection (i.e., future thinking), including affective predictions, is associated with suboptimal decision making and negative mental health and well-being outcomes (Charpentier et al., 2016; Gilbert & Wilson, 2007; MacLeod, 2016; Roepke & Seligman, 2016). However, previous studies have found that prospection can be improved through the use of interventions (Buehler & McFarland, 2001; Roepke & Seligman, 2016). Thus, an intervention specifically targeting affective working memory may be particularly effective in enhancing forecasting accuracy and related processes. This may be best achieved through the use of training, which has been found to improve working memory performance in the cognitive domain (e.g., Soveri, Antfolk, Karlsson, Salo, & Laine, 2017; see also Redick, 2019), and may also be successful in the affective domain (e.g., de Voogd, Wiers, Zwitser, & Salemink, 2016).

Impact Statement

Prospection (e.g., anticipating future outcomes) is a crucial facet of our daily lives, with inaccurate predictions leading to suboptimal decisions, poorer well-being and worse mental health outcomes. This dissertation research focuses on a form of emotional prospection known as affective forecasting and identifies working memory for emotion (i.e., affective working memory) as a core ability underlying how well people can predict their future emotions. These findings suggest that affective working memory may be a promising target for interventions or training programs designed to improve our capacity to maintain emotional states, which can in turn improve people’s ability to forecast their future feelings and make better decisions.