Mood and Judgment in a Dyadic Stress Context

Abstract

The 1980s and 1990s witnessed a surge of psychological research addressing the role of affect in social judgments. Evaluations of others were shown to be shaped, at least in part, by someone’s incidental mood in the moment of social evaluation; while negative moods instigated negative interpersonal evaluations, positive moods instigated positive interpersonal evaluations. Interestingly, these effects only emerged when the reason for the evaluator’s mood was not obvious. Over 30 years later, we expand these findings to the dyadic domain. Rather than conceptualize interpersonal evaluations as occurring solely within an individual, we introduce the Dyadic Affect Infusion/Diffusion (DAID) model to suggest that interpersonal evaluations occur as fundamentally dyadic phenomena. Using six weeks of daily diary data from 311 couples in which one member approached a stressful event, we show that 1) mood influences relationship evaluations at both the intra-individual (i.e. affect infusion) and inter-individual (i.e. affect diffusion) levels, 2) both affect infusion and affect diffusion are turned off by the availability of attributional information, and 3) these effects are better explained by affect infusion/diffusion rather than by several alternative mechanisms. Taken together, the DAID model bolsters the view that individualistic approaches to emotion and social cognition are insufficient and require theory and data at the dyadic and group levels of analysis.

It may seem obvious that our evaluations of other people depend on those people and our relationships with them. However, a surge of psychological research in the 1980s and 1990s called this assumption into question. By randomly assigning participants to experience either positive or negative affective states, researchers showed that someone’s incidental mood in the moment of social evaluation can substantially alter their judgments (Forgas, 1995; Isen, 1987; Schwarz & Clore, 1996). Impression of strangers (Bless, Schwarz, & Kemmelmeier, 1996; Forgas & Bower, 1987), memories of fictitious characters (Mayer, 1992), evaluations of intimate partners (Forgas, Levinger, & Moylan, 1994), and other interpersonal judgments (for reviews, see Forgas, 1995; Schwarz & Clore, 1996) were shown to be shaped, at least in part, by an evaluator’s unrelated and fleeting feelings. This research was exciting not only because it showed the interdependence of ‘feeling’ and ‘thinking’, but also because it had clear implications for real-world settings. Daily life is full of mood inductions—balky public transit, extra work at home, and financial strains—and these events were now implicated in people’s assessments of their social worlds.

Given the negative implications of these findings, researchers were also interested in determining their boundary conditions. In line with Heider’s classic theory on person perception (Heider, 1958), attribution was found to be at least one mitigating factor. Once participants were made aware of the reason for their negative moods, those negative moods no longer shaped how participants evaluated another person (Schwarz & Clore, 1983). This meant that while mood-altering stressful events could negatively impact social judgments, attributing one’s mood to an exogenous cause could prevent this outcome. Researchers during this era therefore provided key insights into the role of affect in interpersonal evaluations, showing not only that mood alters evaluations of an interaction partner, but also that attributions for the mood moderate the effect.

However, while these results seemed to have clear implications for the real world, they were based almost exclusively on studies conducted in controlled lab environments. By design, the only variables that mattered were the participant’s mood and subsequent evaluation of a target person—often a purported but illusory interaction partner. Who the partner was, how they felt, and what they might think of the relationship could not be addressed. But in real world settings, the partner is a distinct individual with their own range of affective states and cognitive processes. In the real world, both people’s affective states and both people’s subsequent evaluations matter.

As will become clear, this seemingly simple idea necessitates a new way of thinking about affect and interpersonal evaluations. The goal of this paper is to articulate this new way of thinking as a novel theoretical model that can be tested in a real-world dyadic context. The model, displayed graphically in Figure 1, allows us to incorporate existing literature on how one person’s affect influences their own interpersonal evaluations, usually called affect infusion. We also introduce new terminology to specify how one person’s affect influences the other person’s interpersonal evaluations, which we call affect diffusion. For each, we discuss how attribution for one’s own and one’s partner’s affect can mitigate these effects. In doing so, we introduce the Dyadic Affect Infusion/Diffusion (DAID) model, which paves the way for future discoveries that are more ecologically valid, sufficiently complex, and comprehensively interpersonal than would be possible by considering any of the model’s components in isolation.

Figure 1.

Theoretical Dyadic Affect Infusion/Diffusion (DAID).

Affect Infusion: AIM

Before explicating the DAID model, we consider existing literature on affect and interpersonal evaluations at the individual level. The process whereby someone’s own affect influences their own social judgments was formalized by the affect infusion model (AIM). According to the AIM, affect influences social judgments intra-individually (hereafter called affect infusion) when people engage in certain cognitive processing strategies typified by their depth of information processing (Forgas, 1992, 1995). Affect infusion occurs when people engage in either heuristic or highly elaborate cognitive processing. While heuristic strategies involve a “how do I feel about them” approach to social evaluation (Schwarz, 1990; Schwarz & Clore, 2003) elaborate processing gives rise to information and memory searches that are inextricably linked with mood (Bower, 1981; Fiedler, 1991; Isen, Shalker, Clark, & Karp, 1978). Importantly, the AIM postulates that attribution only “turns off” affect infusion when heuristic processing is underway (Forgas, 1995).

The AIM has been applied to range of socio-evaluative outcomes and remains the basis for an active area of research. Applied to domains as diverse as perceptions of justice (Mao, Wong, Tao, & Jiang, 2018), decisions about whom to trust (Lowry, Twyman, Pickard, Jenkins, & Bui, 2014), the process of attitude change (Bohner & Dickel, 2011), advice seeking (Gino, Brooks, & Schweitzer, 2012), cooperative behavior (Lin, He, Baruch, & Ashforth, 2017), and more (Beer, 2017; Noval & Stahl, 2017; Piqueras-Fiszman & Jaeger, 2016; Razzaq, Iqbal, Ikramullah, & van Prooijen, 2016; Salerno, 2017; Zheng, Yang, Jin, Qi, & Liu, 2017), the AIM has far-reaching implications and applications.

As depicted in Figure 2, the AIM in its current form can be easily extended to include both people implicated in a socio-evaluative context. In this conceptualization, peoples’ judgements occur in parallel, with each person’s own affect influencing each person’s own interpersonal evaluations. Similarly, each person’s own attributions moderate each person’s own evaluations while engaging in heuristic processing. So, when a heterosexual woman evaluates her intimate partner (Forgas et al., 1994), her incidental mood shapes her evaluation of him, and his incidental mood shapes his evaluation of her, so long as neither knows the reason for their own moods.

Figure 2:

Theoretical affect infusion model (AIM) applied to a dyadic context.

However, such a model does not accord with intuition and research in other domains. Any two people in a dyadic relationship are not independent units but instead form part of a co-regulating dyadic unit, in which each person’s emotions, thoughts, and motivations critically depend on the other (Fitzsimons, Finkel, & vanDellen, 2015). Indeed, one of the most consistent findings in the social psychology of emotion suggests that human emotions are, by nature, interdependent (Butler, 2011; Parkinson, 1996). Variously called affect covariation (e.g. Berg, Wiebe, & Butner, 2011), emotional contagion (e.g. Ruffman, Lorimer, & Scarf, 2017), or emotional transmission (e.g. Frenzel, Becker-Kurz, Pekrun, Goetz, & Lüdtke, 2018), this research shows how people “catch” each other’s mood so that changes in one person’s affect correspond to changes in the other person’s affect. Consequently, the model depicted in Figure 2 should include a left-most arrow linking person A’s affect with person B’s affect. Moreover, the current model leaves unexplored the notion of evaluative covariation. To what extent does person A’s evaluation track with person B’s evaluation? These effects are left unexplored in the current AIM.

Perhaps most importantly, the AIM, as depicted in Figure 2, violates the first principle of interpersonal perception. According to interpersonal perception theory (Kenny, 1994; 2019), social perception differs from object perception in that social stimuli can be reactive and reciprocal; although objects do not judge you back, other people do (see also Heider, 1958). Since interpersonal evaluations are reciprocal, both affect infusion (i.e. an intra-individual effect) and affect diffusion (i.e. an inter-individual effect) must be examined simultaneously. By excluding the possibility of affect diffusion, the current AIM does not control for its influence and is therefore vulnerable to over-estimating intra-individual effects (Cook & Kenny, 2005). But beyond inflated estimates, excluding inter-individual processes produces an incomplete picture of how affect influences interpersonal evaluations in the real world.

Affect Diffusion

Although considerable attention has been paid to affective influences on interpersonal evaluations that occur within a person (i.e. affect infusion), less work has emphasized how one person’s affect impacts another person’s evaluations of them. What terminology should be used for this latter phenomenon? We choose the term diffusion. Merriam Webster defines diffusion as the state of being transmitted. Thus, while infusion pertains to changes that occur within a system, diffusion refers to changes that propagate from one system to another. As such, diffusion is an appropriate term for specifying the process whereby affect in one person influences interpersonal evaluations in another. Note that affect covariation and affect diffusion are similar in that both are interpersonal processes. However, affect covariation concerns one person’s affect influencing another person’s affect whereas affect diffusion refers to one person’s affect influencing another person’s evaluations of them and their relationship. In our view, previous research has examined what we call affect diffusion, but it has done so without a suitable conceptual framework (see Figure 3).

Figure 3:

Theoretical model of existing literature on how one person’s affect impacts another person’s interpersonal evaluations—a phenomenon we call affect diffusion.

For example, prior research shows that confederates who express feelings of happiness, sadness, or irritability are liked more than those who express no emotions (Clark & Taraban, 1991). Moreover, people who express their emotions tend to have more friends and report higher levels of intimacy with those friends relative to their less expressive counterparts (Graham, Huang, Clark, & Helgeson, 2008). Although these effects are stronger in communal (rather than exchange) relationships (Clark, Ouellette, Powell, & Millberg, 1987), converging evidence suggests that emotional expression in one person influences how the other person evaluates them (Parkinson, 1996; Peters & Jameison, 2016; Peters, Overall, & Jameison, 2014; Stellar et al., 2017).

Of course, the mechanisms that explain affect diffusion likely differ from those that explain affect infusion. Thus, while the process of affect infusion is clearly articulated in the original AIM, we must turn to other literature to understand the process of affect diffusion. The Emotions as Social Information Model provides a good basis, as it articulates how emotions evolved to communicate relevant information from one person to the next (Van Kleef, 2009). This is because negative emotions signal that something is wrong and requires reconciliation (Fischer & Manstead, 2008; Graham, Huang, Clark, & Helgeson, 2008; Hareli & Hess, 2010; Lemay, Overall, & Clark, 2012), although the efficacy of such emotional expression depends on the attachment style of the perceiver (Overall, Gletcher, Simpson, & Filo, 2015). The fact that one person’s emotional expression corresponds to another person’s inferences about how genuine that person is (Krumhuber & Manstead, 2009), the value of that person’s clothes (Peace, Miles, & Johnston, 2006), and the quality of that person’s help (Söderlund & Rosengren, 2010) further corroborates the idea that emotions serve important social functions. Perhaps affect infusion occurs because people ask themselves “how do I feel about them” (Schwarz & Clore, 1983) while affect diffusion occurs because how “they” feel signals important information for “me” and “us.”

Interestingly, the stress literature also elucidates affect diffusion processes, as the number and severity of stressors in one person’s life correspond to their close partner’s evaluations of them and their relationship. Termed stress crossover, this research shows that stressors experienced by one partner decrease the other partner’s relationship satisfaction (Lavner & Clark, 2017; Neff & Karney, 2017), attitudes toward the stressed partner (Crouter, Bumpus, Head, & McHale, 2001), as well as behaviors towards them at the end of the day (Schulz, Cowan, Pape Cowan, & Brennan, 2004). Since stressors are known to disrupt mood (Bolger, DeLongis, Kessler, & Schilling, 1989; Giles, Mahoney, Brunyé, Taylor, & Kanarek, 2014), these findings imply that stressors sometimes have negative relationship consequences because of their impact on mood. Theoretically, these effects occur via daily interactions that undermine relationship quality (Neff & Karney, 2009) and depend on gender (Neff & Karney, 2007), socioeconomic context (Crown & Karney, 2007), and expectations about the relationship (McNulty & Karney, 2004).

Indeed, one study in the stress literature directly tested affect diffusion in daily life—both its existence and its moderation by attributional salience. In this diary study, one partner reported their mood every day for four weeks leading up to a stressful event while the other partner provided daily evaluations of their relationship (Thompson & Bolger, 1999). Results showed that higher levels of depression in the stressed partner corresponded to lower levels of relationship happiness in the other partner, both on the same day and in cross-lagged models. Importantly, this study also found that the effect diminished as the reason for the stressed partner’s negative mood became increasingly salient: As the stressful event approached, the partner’s relationship evaluations were less impacted by the stressed partner’s negative mood. As such, affect diffusion decreased as attributional salience increased, which substantiates affect diffusion as depicted in Figure 3.

Nonetheless, the model depicted in Figure 3 suffers the same limitations as those in the affect infusion literature. Both affect covariation and evaluative covariation have been largely ignored in these prior studies, and most of the time affective influences on social evaluations at the intra-individual level (i.e. affect infusion) have been excluded. As a result, affect diffusion may be over-estimated in some of this previous work (Cook & Kenny, 2005). Most importantly, without jointly considering affect infusion and affect diffusion, it is actually impossible to tell whether these findings truly represent a diffusion process, versus an infusion process that was artificially excluded. We discuss this possibility next.

Dyadic Affect Infusion/Diffusion: DAID

In light of these parallel literatures, one might be tempted to conclude that both affect infusion and affect diffusion shape interpersonal evaluations. However, by failing to conceptualize and model both processes simultaneously, the relative importance—and existence—of each is unclear. This is because affect covaries between partners in a way that obfuscates which process actually transpires. This issue is best understood through an illustrative example with two people who we will call Lyn and Randall.

In studies documenting affect infusion, it appears that each person’s own mood impacts each person’s own assessment of their relationship and the other person. As suggested by prior literature, Lyn’s negative mood could lead Lyn to judge her relationship with Randall negatively. Furthermore, since affect covaries between relationship partners, Lyn’s negative mood predicts Randall’s negative mood. Because of affect infusion, Randall’s negative mood could also lead him to judge the relationship negatively. In this case, affect covariation leads to negative evaluations in both partners primarily by affect infusion processes. However, without also modelling affect diffusion, it is impossible to tell whether this is solely the case. It is also possible that Lyn’s negative mood leads Randall to judge the relationship negatively via diffusion. Since affect covaries, Lyn would not be the only one in a negative mood: Randall would also be in a negative mood. Because of affect diffusion, Randall’s negative mood could lead Lyn to evaluate the relationship negatively. In this scenario, the same effect emerges (i.e. negative mood in both partners and negative evaluations in both partners) but can be explained primarily by affect diffusion rather than by affect infusion. The fact that affect covaries between people means that both processes must be examined at the same time—otherwise it is impossible to tell which process, or both, occurs.

The DAID model addresses this issue by jointly considering affect infusion and diffusion. The principal contribution of the DAID model is therefore its ability to estimate affect infusion while controlling for affect diffusion and vice versa, as well as to estimate affect covariation and evaluative covariation. But DAID also offers a number of other insights. First, DAID more closely approximates real-world processes by refusing to ignore the person being evaluated in any social context, an emphasis that has been criticized as missing from the existing literature on affect infusion (see Clark, 2002). Second, DAID introduces dyadic, rather than individualistic, mechanisms linking affect to interpersonal evaluations. Specifically, one person’s affect can lead another person to evaluate their relationship differently because of affect covariation. This would happen when Lyn’s affect influences Randall’s affect, which in turn shapes how Randall evaluates the relationship—in other words affect diffusion can be explained by affect covariation and the same could be true for affect infusion.

DAID also contributes to the burgeoning literature on emotional accuracy/projection, which makes use of a model very similar to the DAID model, minus the postulate of moderation by attributional salience. In the accuracy/projection literature, the left-most constructs in Figure 1 are person A and person B’s emotions, while the right-most constructs are person A’s perception of person B’s emotions and person B’s perception of person A’s emotions. Returning to our example with Lyn and Randall, this literature finds that Lyn’s perception of Randall’s emotions depend partly on how Randall feels (accuracy) and partly on how Lyn feels (projection; Chang, Overall, Madden, & Low, 2018; Clark, Von Culin, Clark-Poner, & Lemay, 2017; Overall, Clark, Fletcher, Peters, & Chang, 2019). These findings are related to DAID insofar as they constitute an earlier stage of the person perception process. According to the Stage Model of Person Perception (DeVito & DeVito, 2007), perceiving others’ emotions occurs during the second stage, Organization, while evaluating other people and their emotions occurs during the third stage, Interpretation-Evaluation. It may therefore be the case that emotional projection/accuracy partially mediates the relation between emotional experiences and interpersonal evaluations, as evidenced by one study showing that expressing emotional vulnerabilities to one’s partner leads to beliefs that the partner views them as vulnerable (projection), which in turn causes them to evaluate the relationship negatively (infusion; Lemay & Clark, 2008).

Since DAID concerns how affect influences interpersonal evaluations when affect covaries between people, and since affect covaries between many kinds of dyads—even between newly acquainted ones (Bruder, Dosmukhambetova, Nerb, & Manstea, 2012; Cwir, Carr, Walton, & Spencer, 2011; Scarpa et al., 2018)— the model could be tested in a variety of contexts. Next, we examine the first empirical test of the DAID model with a naturalistic experiment for couples undergoing a highly stressful event.

Current Study: Empirical Investigation of DAID

To empirically test the DAID model, we operationalized negative affect as daily levels of depression, as defined by the Profile of Mood States, and interpersonal evaluations as daily evaluations of a romantic partner. We employed a naturalistic daily diary experiment with the same design as Thompson & Bolger (1999), in which couples provided daily reports while one partner underwent a stressful event. Specifically, one member of the couple prepared for, finished, and then moved on from taking the New York Bar Exam. The bar exam represents the culmination of years of hard work and considerable financial investment, and subsequently evokes high levels of distress for those who take it. The examinee, the partner, and the relationship were increasingly strained in the time period leading up the bar exam (Shrout et al., 2010), although the reason for only one partner’s negative mood (the examinee) became increasingly obvious as the exam approached.

This was, therefore, an intensive prospective design that obtained daily reports of mood and relationship evaluations from both members of the dyad in the weeks leading up to and including the bar exam, as well as one week following it. This design had several features that simplified the investigation of DAID. First, the fact that the bar exam was a scheduled event allowed us to examine affect infusion/diffusion prospectively in the weeks leading up to and after the exam. We followed couples from a time when the reason for the examinee’s negative mood was not obvious to a time when a reason for the examinee’s mood was readily available. Moreover, this clear attribution for the examinee’s mood was reduced immediately after the exam ended. Second, we recruited couples where only one member underwent the stressful event, which standardized the reason for the examinee’s mood and provided no clear attribution for the partner’s mood. Third, the context in which the data were collected allowed us to investigate the DAID model in a real-life setting with real-world relationships.

Following prior work (Thompson & Bolger, 1999), we focused on how daily levels of depression among both the examinee and the partner influenced each member’s negative evaluations of the relationship.¹ We hypothesized that at the start of the study, when the exam was too distal to provide a clear attribution for the examinee’s mood, both partners’ negative mood would correspond with negative evaluations for both the examinee and the partner. This would support a main effect of affect infusion/diffusion such that the examinee’s mood would impact the examinee’s evaluations (affect infusion) and the partner’s evaluations (affect diffusion). Similarly, at the start of the diary, the partner’s mood should impact both their own evaluations (affect infusion) and the examinee’s evaluations (affect diffusion). However, we also predicted that as the exam approached, the reason for the examinee’s negative mood would become clear so that the slope between the examinee’s mood and the examinee’s relationship evaluation should decrease (i.e. moderation of affect infusion by attributional salience). At the same time, the slope between the examinee’s mood and the partner’s relationship evaluations should also decrease, as the partner would also have an attribution for the examinee’s mood (i.e. moderation of affect diffusion by attributional salience). We made no such moderation predictions for the mood of the partner: as the exam approached, the reason for the partner’s mood remained relatively constant. Therefore, the association between the partner’s mood and both the partner’s (affect infusion) and the examinee’s (affect diffusion) evaluations should remain the same. We made similar predictions for the time immediately after the exam. Once the exam was over, attributions for the examinee’s mood were likely less clear. As such, the relation between the examinee’s mood and the examinee’s and the partner’s relationship evaluations should return to their initial levels. These hypotheses are depicted visually in Figure 4. All code and analyses that follow can be found at: https://osf.io/dgw29/?view_only=ece960c31743417b9e6e718a27aa09e6.

Figure 4.

Hypothesized DAID model. Solid lines represent affect infusion and dashed lines represent affect diffusion. Note that all effects on the left are correlated with one another.

Methods

Sample

We recruited couples in long-term, intimate, heterosexual relationships in which one partner would take the bar exam. More than 100 law schools were contacted in the continental United States in the summers of 2001, 2002, and 2003 and ultimately reached over 15,000 students. Eligible participants and their partners were married or cohabiting for at least 6 months at the time of recruitment and only one member of the couple planned on taking the exam. Of 765 eligible couples, 552 were assigned to the diary condition. The other group of eligible couples was assigned to a cross-sectional survey condition that will not be considered here. Of those in the diary condition, 472 (77%) couples agreed to participate, 81 of whom were asked to participate in a different version of the diary study, and the remaining 391 of whom were asked to participate in the version reported here (see Shrout et al., 2018 for further details on study design). Eighty of those couples had at least one member who did not respond to at least one of the measures required to run the model, so 311 couples were ultimately included in the final analysis. Although the sample size was based on a power analysis to address a separate research question (see Shrout et al., 2010), the data contain considerable repeated measurements from a large number of couples and therefore provide a reasonable test of the DAID model.

Couples were paid $150 for participation, with the chance to win $1,000 if both members of the couple completed the study. Couples received an initial payment of $10, two consent forms, two background questionnaires, and two return envelopes. The diary period took place in the five weeks leading up to the exam, the two days on which the exam was taken, and one week after the exam. Participants received packets of diaries by mail on a weekly basis (six packets over the six weeks of the study). Each batch consisted of seven identically structured daily diaries, with the exception of the last batch, which consisted of nine daily diaries. The diary form included a morning section and evening section. The morning section was brief and contained a subset of items from the evening diary. The diary had a range of questions regarding mood, feelings about the relationship, daily troubles or difficulties, relationship conflicts, how much time they had spent together in the previous 24 hours, and support transactions. Participants were asked to complete the questionnaires separately and not to share or discuss their answers with one another. Participants were also asked to complete the diaries on the days assigned and 88% of the completed diaries were reported to be completed on the correct day. Due to the method of recruiting participants, these participants might not be fully representative of those preparing for the bar exam.

On average, examinees were 30 years old (SD = 6) and partners were 30 years old (SD = 8). Fifty-four percent of the examinees were female and 65% of the couples were married. The racial/ethnic composition of the examinees was 80% White, 7% Asian, 2% Black, 5% Latinx, and 5% Other and the partners were 76% White, 5% Asian, 4% Black, 5% Latinx, and 10% Other. Partners were typically highly educated, with only 17% of the partners having less than a 4-year college degree and 80% of those with less than a 4-year degree having some college education.

Measures

Time.

Time was operationalized as beginning on 0 and ending on day 43. The exam took place over the course of two days, on the 35^th and 36^th day. We recoded time so that it incremented by one each day before the exam ended (days 0–35) and remained constant once the exam was over (i.e. days 36–43 were recoded to be 36). This recoding implies a changing effect of time before the exam and a constant effect of time once the exam was over. We did this because there was no clear theoretical reason to expect the effect of mood on relationship evaluations to vary as a function of time after the exam; while the time before the exam tracked the salience of attributional information (i.e. attributional salience increased as the exam approached), time after the exam did not have a clear meaning. We then divided the time variable by 30 so that its units were in months. The time coefficient can therefore be interpreted as the change in relationship feelings in the month leading up to the exam.

Exam Finished.

We created a dichotomous variable to specify that the exam ended. This variable was 0 for every day leading up to the exam (days 0–35) and 1 for each day following the exam (days 36–43).²

Daily Index of Affect: Depression.

In line with Thompson and Bolger (1999), we used a composite score of depressed mood that was derived from the POMS (Norcross, Guadagnoli, & Prochaska, 1984). Participants rated how discouraged, blue, sad, and hopeless they felt on a scale from 0=not at all to 4=extremely in both the morning and evening section of the diary. Items were averaged and the variable was rescaled to range from 0 to 10 for ease of interpretation.

Daily Index of Negative Relationship Evaluations.

We created a composite measure of negative relationship evaluations. Items were generated as an extension of previous scale development on the structure of cognitive representations of intimate relationships (see Fletcher, Simpson, & Thomas, 2000). In the evening section of the diary, participants indicated on a scale from 0=not at all to 4=extremely how fearful, sad, angry, irritated, worried, and depressed they felt about their relationship. Since participants indicated how they felt about their relationship, the measure is fundamentally evaluative in nature. Rather than report on a current feeling (i.e. how do you feel right now), they reported on an evaluation (i.e. how do you feel about the relationship right now). These items were averaged and rescaled to range from 0 to 10 for ease of interpretation.

Time Spent Together.

Each evening, participants also reported how much time they had spent together in the previous 24 hours, not including sleeping. Examinee and partner reports of time spent together were highly correlated (r=0.74[0.73, 0.75]), so we took their average. This variable was used in a follow-up analysis.

Statistical Analysis

The DAID model requires that both members of the dyad be conceptualized—and therefore modelled—as part of a dependent unit. The Actor Partner Interdependence Model (APIM) is a statistical technique that recognizes interdependence between people in a relationship by allowing observations among dyads to be correlated (Cook & Kenny, 2005), and Figure 4 above is a SEM-style depiction (see Hoyle, 2012). In an APIM, dependent and independent variables for each dyad member covary, which gives rise to the specification of actor and partner effects. While actor effects refer to how a variable for person A influences an outcome for person A, partner effects refer to how a variable for person A influences an outcome for person B. Importantly, actor effects are estimated simultaneously with partner effects, and therefore reveals the extent to which each process (e.g. actor effects) occurs above and beyond the other (e.g. partner effects). Applied to the DAID model, an APIM enables relationship partners’ moods to covary—a well-documented phenomenon (for a review, see Butler & Randall, 2013)—and each partner’s mood to predict both their own and their partner’s evaluations of the relationship. Lastly, this technique quantifies residual covariation in relationship evaluations, a phenomenon that has received surprisingly little attention.

What is not obvious from the model depiction in Figure 4 is that each effect parameter is allowed to vary across dyads. Rather than simply estimate how mood influences relationship evaluations on average (termed fixed effects in the language of multilevel models), the models also estimate how much couples differ from one another in this process (termed random effects). In order to simplify an already complicated discussion of mood effects on relationship judgments, we omit interpretations of random effects in the results section below. Although a deep discussion of random effects is beyond the scope of the current paper, we discuss them in the Supplemental Materials. Readers interested in between-dyad heterogeneity should therefore refer to these Supplemental Materials as well as other related articles (i.e. Bolger, Zee, Rossignac-Milon, & Hassin, 2019).

All models were estimated with Bayesian techniques. Bayesian approaches are increasingly popular in the psychological sciences (Van De Schoot, Winter, Ryan, Zondervan-Zwijnenburg, & Depaoli, 2017), and when employed with non-informative priors, give rise to very similar estimates as models using a frequentist perspective. A benefit of Bayesian models is that they allow us to make direct probability statements about plausible values of model parameters. This is not possible in frequentist approaches, which result in probability statements about the likelihood of data values rather than parameter values (Gelman, Dunson, Rubin, Stern, & Carlin, 2013; McElreath, 2018). Also, Bayesian estimation provides the entire probability distribution of parameters that are consistent with the sample data, and in presenting results, we visualize all effects as posterior probability distributions. This approach aligns with rising concerns regarding the shortcomings of binary significance testing (Dienes, 2011; Wagenmakers, 2007).

Results

We now turn to the results, where we examine the effect of mood on negative relationship evaluations as well as alternative models that address potential confounding factors.

First, we test affect covariation with an intercept only model in which the residual correlation reveals the extent to which one partner’s mood correlates with the other partner’s mood. Second, we test the remaining DAID hypotheses with the model depicted in Figure 4. In this model, we test main effects of affect infusion and affect diffusion by looking at the intercepts for mood in these models: these parameters show the extent to which mood at the beginning of the diary correlates with relationship evaluations at the beginning of the diary. Third, we test whether these main effects are moderated by attribution by investigating the interaction between mood and time (because time indexes increasing attributional salience) and between mood and the exam being over (because exam completion indexes decreased attributional salience). These final parameters reveal whether attributional salience influences affect infusion and affect diffusion for the examinee but not the partner. Fourth, we test evaluative covariation via residual correlations in the model. This parameter quantifies the extent to which residuals in relationship evaluations for one partner covary with residual relationship evaluations in the other partner. Finally, we note that this model includes estimates for stress spillover³ and stress crossover;⁴ these are seen in the main effects of time and main effects of exam completion. Because these effects do not concern DAID, we include them only in the Supplemental Materials.

Primary Analysis

Average levels of depressed mood for both the examinee and the partner can be found in Figure 5.

Figure 5:

Examinee and partner daily depression. Vertical dashed lines represent exam days.

Covarying Variables: Affect Covariation and Residual Evaluative Covariation.

Results can be found in Table 1 and Figure 6. Affect covariation was estimated via the intercept only model discussed above, and results show a positive relation between the examinee and the partner’s moods (r=0.12[0.10, 0.14]). Similarly, negative relationship evaluations covaried within-couple, such that residual negativity in one partner’s relationship evaluations correlated with residual negativity in the other partner’s relationship evaluations (r=0.13[0.11, 0.15]).

Table 1:

Multilevel APIM Results for Examinee and Partner Negative Relationship Evaluations Regressed on Examinee and Partner Depressed Mood – Average (Fixed) Effects

	Examinee’s Negative Evaluations	Partner’s Negative Evaluations
Examinee NM (affect infusion & diffusion)	0.33 [0.28, 0.38]*	0.08 [0.06, 0.11]*
Examinee NM * Time	−0.15 [−0.21, −0.09]*	−0.04 [−0.07, −0.003]*
(modulation of affect infusion & diffusion)
Examinee NM * Finished (modulation of affect infusion & diffusion)	0.05 [0.02, 0.09]*	0.03 [0.0005, 0.07]*
Partner NM (affect infusion & diffusion)	0.10 [0.05, 0.15]*	0.43 [0.37, 0.49]*
Partner NM * Time (modulation of affect infusion & diffusion)	0.04 [−0.03, 0.11]	−0.02 [−0.11, 0.07]
Partner NM * Finished (modulation of affect infusion & diffusion)	0.02 [−0.04, 0.07]	0.09 [0.04, 0.14]*
Affect Covariation	0.12 [0.10, 0.14]*
Evaluative Covariation	0.13 [0.11, 0.15]*

NM=Negative mood operationalized as depression. CI=Bayesian credibility interval. Asterisks flag effects that are estimated to be non-zero in the 95% credibility interval.

Figure 6:

Affect infusion/diffusion. Blue lines are draws from the posterior distributions of effects for the examinees’ negative mood while red lines are draws from the partners’. Vertical lines represent exam days. Note distinct y-axes per graph.

Actor Effects: Affect Infusion.

The actor effects in the APIM align with existing work in the domain of affect infusion: the ‘actor side’ reveals the extent to which one’s own mood corresponds to negative evaluations of one’s intimate relationship. At the beginning of the diary (i.e. the y-intercepts in Figure 6), both examinees (b=0.33[0.28, 0.38]) and partners (b=0.43[0.37, 0.49]) exhibited affect infusion such that higher levels of one’s own negative mood correlated with more negativity about the relationship. For examinees, being one unit higher in negative mood corresponded with being 1/3 of a unit higher in negative relationship evaluations, while for the partner, being one unit higher in negative mood corresponded with being 2/5 of a unit higher on negative evaluations. Since these effects were reliably non-zero, they replicate and extend prior research on affect infusion by demonstrating that affect infusion takes place in daily life and when controlling for affect diffusion.

Although both examinees and partners exhibited affect infusion at the start of the diary, the effect of one’s own mood on one’s own perception of the relationship over time varied between examinees and partners. For examinees, affect infusion decreased over time, such that the gradual approach of the exam reduced the magnitude of the slope between mood and relationship evaluations (b=−0.15[−0.21, −0.09]). While being one unit higher in negative mood corresponded with being 1/3 of a unit higher in negative evaluations at the beginning of the diary, by the day of the exam, being one unit higher in negative mood only corresponded with being 1/10 of a unit higher in negative relationship evaluations. In other words, the magnitude of the slope decreased over time. Once the exam was over, mood again had a reliable effect on relationship evaluations (b=0.05[0.02, 0.09])—back up to 2/5 of a unit. We can infer that the reason mood played a decreasing role in negative relationship evaluations was because the examinee had an increasing external attribution for their own mood as time passed. As such, examinees discounted the effects of their mood on their relationship evaluations. However, for partners, time did not reliably moderate the effect of mood on relationship judgments (b=−0.02[−0.11, 0.07]). Since the partner had no clear reason for their negative mood, the increasing approach of the exam did not moderate these mood effects. Finally, and contrary to our predictions, the end of the exam did moderate the effect of the partners’ mood on their own relationship evaluations (b=0.09[0.04, 0.14]).

It is important to recognize that these affect infusion effects occurred above and beyond affect diffusion. Prior to analyzing this data, it was conceivable that covariation in negative mood would outweigh affect infusion; it could have been that as examinees became increasingly depressed, their partner also became increasingly depressed (due to affect covariation), which led each to judge the relationship negatively due to the other partner’s mood. Rather than have one’s own mood influence one’s own judgments, increases in relationship negativity could have been driven primarily by the other partner being in an increasingly bad mood. The fact that affect infusion occurred above and beyond affect diffusion is a testament to the strength of affect infusion in daily life. The extent to which the alternative process occurs, in which one partner’s mood influences the other partner’s relationship assessment above and beyond affect infusion, is seen in the partner side of the APIM.

Partner Effects: Affect Diffusion.

The partner effects outlined here shed light on affect diffusion: the ‘partner side’ of the APIM reveals the extent to which one partner’s mood influences the other partner’s evaluations of the relationship. At the beginning of the diary period, both examinees (b=0.10[0.05, 0.15]) and partners (b=0.08[0.06, 0.11]) displayed affect diffusion such that higher levels of one’s own negative mood correlated with greater levels of the other partner’s negativity about the relationship. For examinees, being one unit higher in negative mood corresponded with partners being about 1/10 of a unit higher in negative relationship evaluations, and the effect was the same for the partner’s mood. These effects were reliably estimated to be non-zero.

Although both examinees and partners exhibited affect diffusion at the start of the diary, the way each partner’s mood impacted the other partner’s negative relationship judgments over time depended on which member of the dyad was in a depressed mood. For partners, the effect of the examinee’s mood on their partner’s judgments about the relationship decreased over time, such that the approaching exam reduced the magnitude of affect diffusion (b=−0.04[−0.07, −0.003]). Although the examinee being one unit higher in negative mood corresponded with the partner being 1/10 of a unit higher in negative relationship evaluations at the start of the diary, by the day of the exam the effect was down to 3/100 of a unit. Once the exam was over, the examinee’s negative mood once again influenced how negatively the partner viewed the relationship (b=0.03[0.0005, 0.07]), back up to 1/10 of a unit. As such, partners’ relationship assessments became decreasingly dependent on the examinee’s mood, likely because the reason for the examinee’s negative mood became increasingly obvious over time. However, for examinees, time did not reliably moderate the effect of their partner’s mood on the examinees’ negative relationship judgments (b=0.04[−0.03, 0.10]). This likely occurred due to the lack of attributional salience for the partners’ negative mood. Moreover, the effect of the partner’s mood on the examinee’s negative feelings about the relationship did not reliably increase in the time immediately following the exam (b=0.02[−0.04, 0.07]).

Results Summary

These empirical results support the hypothesized model in Figure 4 with one exception. As predicted, affect infusion and diffusion occurred for both partners and examinees at the start of the diary. Also as predicted, the effect of the examinee’s mood on both the examinee’s and the partner’s relationship evaluations was moderated by time and the exam ending: the examinee’s negative mood played a decreasing role in their own and their partner’s relationship evaluations over time, and when the exam ended the examinee’s negative mood once again influenced their own and their partner’s judgments. Additionally, we predicted that the effect of the partner’s negative mood on both the partner’s and examinee’s relationship evaluations would not be reliably moderated by time or the exam being over. This was true of all effects except that of the partner’s negative mood on the partner’s relationship evaluations once the exam was over. We discuss why this likely occurred in one of the ensuing follow-up analyses.

Follow-Up Analyses

Although the context of the study supports the intuitive and theory-driven hypotheses of the DAID model, there are alternative explanations that could confound the interpretations made thus far. Tables for these results can be found in Supplemental Materials.

Reverse Causation Explanation.

The central effect discussed, in which negative mood dampened relationship evaluations, could be explained by a reverse causal process: Couples in this sample may have felt increasingly worse about their relationships, which led them to be in worse moods.

We therefore tested a prospective model to confirm that morning negative mood predicted that evening’s relationship evaluations, controlling for the previous evening’s relationship evaluations (relationship evaluations were not collected in the morning, so we could not control for morning relationship evaluations). For simplicity, we specified these APIMs as non-varying over time. The results from these analyses reveal that both examinees’ and partners’ morning mood were associated with evening relationship evaluations. On the actor side of the model, both examinee’s and partner’s own depressed mood predicted more negative relationship evaluations (examinee: b=0.13[0.10, 0.16]; partner: b=0.12[0.09, 0.16]). On the partner side of the model, we saw conceptually similar results. Each partner’s depressed mood in the morning played an explanatory role in evening relationship evaluations, such that high levels of morning depressed mood corresponded to higher levels of evening relationship negativity (examinee: b=0.03[0.01, 0.05]; partner: b=0.02[−0.01, 0.05]). Although one effect (the partner’s morning mood on the examinee’s evening evaluations) was not statistically reliable, it was in the expected direction. It seems that the general pattern of results discussed above, in which depressed mood predicted relationship evaluations, was warranted since morning negative mood generally predicted evening relationship evaluations controlling for last night’s relationship evaluations.

Withdrawal Explanation.

Next, it is possible that the moderation effect, whereby negative mood played a decreasing role on both examinee’s and partner’s assessments of the relationship, happened not because of attributional salience but because of social withdrawal in the dyad. According to the vulnerability-stress-adaptation model, stressful events cause couples to withdraw from the relationship either because they have less time to engage in shared activities or because they do not want to expose their diminished intraindividual resources to the demands of the relationship (Karney & Bradbury, 1995; Neff & Karney, 2009). A more positive interpretation of the withdrawal phenomenon conceptualizes withdrawal as a constructive process that shields the relationship from the pernicious effects of stress (Bolger, DeLongis, Kessler, & Wethington, 1989; Repetti, 1989). In either case, prior research suggests that couples withdraw from relationships during stressful times. It is therefore possible that couples in this sample withdrew from the relationship as the exam approached, and that affect decreasingly impacted relationship evaluations because couples were spending less time together. Indeed, in a secondary analysis regressing average reported hours spent together on the time and exam finished variables found that couples spent less time together leading up to the exam (b=−1.32[−1.54, −1.11]) and more time together once the exam was over (b=3.34[3.11, 3.57). These findings replicate prior research showing that couples withdraw under stress.

To test whether withdrawal explained infusion/diffusion, we re-ran the primary APIM model discussed above, but this time included withdrawal as a co-variate (with withdrawal operationalized as the average of both partner’s reported hours spent together). This analysis reveals whether affect infusion/diffusion occurred above and beyond withdrawal. The results remained overwhelmingly consistent, with all but one effect in the original model replicating. The only result that did not replicate while controlling for withdrawal was moderation of affect diffusion for the examinee’s mood on the partner’s negative evaluation of the relationship. Since every other effect replicated, these results suggest that DAID occurred for reasons other than withdrawal.

Discussion

“In dealing with the person as a member of a dyad, he cannot be described as a lone subject in an impersonal environment, but must be represented as standing in relation to and interacting with another person” (Heider, 1958, p. 1)

We sought to heed Heider’s advice by building a theoretical and empirical model of affective influences on interpersonal evaluations in the smallest truly social unit, the dyad. We proposed the Dyadic Affect Infusion/Diffusion (DAID) model, which was based on existing literature suggesting that affect impacts interpersonal evaluations at both the intra-individual level (what is typically called affect infusion) and inter-individual level (what we termed affect diffusion). Theoretically, we postulated that only by conceptualizing and modelling these phenomena simultaneously could we gain a full understanding of their effects. We tested the DAID model with a naturalistic experiment in which couples underwent a highly stressful event over the course of six weeks.

Replicating prior work conducted at the intra-individual level (Forgas et al., 1994; Tesser & Beach, 1998), we found that affect infusion impacted close relationship evaluations, such that each partner’s own mood influenced each partner’s own relationship evaluations. This effect was found above and beyond the effect of each partner’s mood on the other partner’s evaluations. At the same time, we also replicated prior research on affect diffusion (Thompson & Bolger, 1999), such that each partner’s own mood influenced the other partner’s relationship evaluations. We extended these prior results by showing these effects above and beyond affect infusion. Finally, addressing calls in the close relationships literature to better understand factors that may buffer relationships from the negative ramifications of stress (Story & Bradbury, 2004), we found that attributional salience dampened affect infusion/diffusion: As the stressful event approached, both the examinee and the partner were less impacted by the examinee’s mood. Such moderation emerged less reliably for the partner’s mood, which makes sense given the lack of an attribution for the partner’s mood. We ruled out two alternative explanations for these effects by showing 1) the causal arrows implicated in our model were grounded in prospective reports of mood and relationship evaluations and 2) the dampening of affect infusion/diffusion over time mostly persisted in models controlling for withdrawal processes. Taken together, these findings support the hypothesized DAID model, which has several implications for research on affect and social cognition, on close relationships, and on stress.

Implications for Affect and Social Cognition

Researchers have long denounced the notion that affect and cognition constitute separate research domains (e.g, Zajonc, 2000), arguing instead that affect plays a critical role in explaining how people think about their social worlds. Although an extensive body of experimental research now corroborates this theory at the intra-individual level (for reviews, see Forgas, 1995; Schwarz & Clore, 1996, 2003) fewer studies have generalized these findings beyond the lab. Our daily diary data demonstrate that affect infusion impacts interpersonal evaluations among real-world dyads undergoing a real-life stressful event.

The results also imply that existing research on affect infusion can be expanded beyond the individual level and taken into the dyadic domain. Mood covaried between romantic partners over time, which not only affected each partner’s own assessments of the relationship, but also the other partner’s assessments. This finding is important because it acknowledges the social origins of mood; affect does not remain within an individual nor does it shape only individual cognition. Rather, affective experiences are dyadic and have dyadic consequences.

Such a finding is especially relevant given recent advances in Heiderian theorizing. Although Heider’s balance theory and attribution theory have typically been conceptualized separately, a recently uncovered trove of his personal notebooks and journals show that he later developed a unified theory of the two (Crandall et al., 2007). These notebooks reveal a cohesive Heiderian view on balance and attribution, suggesting that people make attributions for others as a way to maintain a balanced viewpoint of them. In the context of close relationships, people are motivated to view their partner in a positive light (Murray & Holmes, 1997; Murray, Holmes, Dolderman, & Griffin, 2000; Rusbult, Van Lange, Wildschut, Yovetich, & Verette, 2000; Song et al., 2018), the corollary being that they should actively seek attributional information in order to maintain positivity about their relationships. Our results support this theoretical supposition, as increasing attributional salience for someone’s own (in the examinee’s case) and someone’s partner’s (in the partner’s case) mood dampened affect infusion/diffusion.

This Heiderian perspective has further implications for the DAID model that offer an exciting avenue for future research; namely whether these effects persist for positive moods. As discussed elsewhere (Isen, 2002), positive affect influences social cognition intra-individually because positive emotions and mood promote more flexible, creative, and adaptive cognitions that can result in more positive evaluations of other people. Moreover, affect diffusion has been documented for positive affect, as people tend to like others who express happiness (see Van Kleef, De Dreu, & Manstead, 2010). Given our findings that affect infusion persisted while controlling for affect diffusion and vice versa for negative moods, we suspect that simultaneous affect infusion/diffusion similarly persists for positive moods. At the same time, Heiderian balance implies that people may ignore attributional information for positive moods; affect infusion/diffusion might not be “turned off” for positive moods because doing so would create a mismatch between being in a relationship with someone and thinking positively of them. As such, future research on DAID could test whether attributional salience fails to “turn off” affect infusion/diffusion for positive affect.

As another contribution to the affect infusion literature, our data suggest that even evaluations of a close partner are vulnerable to affect infusion/diffusion. Previously, evaluations of close relationships were considered part of substantive processing, meaning that they should be unaffected by attribution processes since attribution only moderates affect infusion for heuristic processing (Forgas, 1992; Forgas et al., 1994). In contrast, our results suggest that daily evaluations of a partner are moderated by attribution. This implies that even evaluations of a romantic partner can be governed by heuristic processing. The affect infusion literature should acknowledge that heuristic processing, and subsequently moderation by attributional salience, is not necessarily a function of who is being evaluated, but rather in what context (i.e. daily) and for what question type (i.e. momentary).

Taken together, these results contribute to emerging research on emotional accuracy/projection. As mentioned earlier, emotional projection is conceptually similar to affect infusion and emotional accuracy is conceptually similar to affect diffusion, although both accuracy and projection concern perceiving emotions rather than making evaluations based on those emotions. An exciting next step could be to examine the reciprocal relation between these effects, since perception both precedes and follows from evaluation (DeVito & DeVito, 2007). Do people who accurately perceive their partner’s emotions go on to evaluate their relationships differently because of those emotions (diffusion), and do those altered interpersonal evaluations in turn determine the ability to perceive how one’s partner feels? And do these reciprocal relations imply that attributional information also moderates emotional accuracy/projection? For example, people might refrain from projecting their own negative emotional states onto their partners when they are aware of the reason for those emotions. Bridging the literature on emotional accuracy/projection with that on affect infusion (and now diffusion) raises intriguing questions that can be addressed in future research.

Implications for Dyadic Relationships and Stress

The DAID model has further implications for the literature on close relationships and stress, particularly our findings on affect diffusion. Theorists have outlined two primary mechanisms to explain why stress causes relationship deterioration. The first is that stress creates additional marital problems and the second is that stress hinders spouses’ ability to respond to those problems constructively (Neff & Karney, 2009, 2017). Our affect diffusion findings suggest that both mechanisms may be explained by the negative mood instigated by stressors; i.e. stressors cause negative moods which increase marital problems and decrease the ability to deal with such problems. In line with the former, one study found that workplace stress leads to negative mood in the stressed partner, which then leads to hostile interpersonal interactions (Lim et al., 2018). In line with the latter, stress is known to instigate negative mood (Bolger, DeLongis, Kessler, & Schilling, 1989; Giles, Mahoney, Brunyé, Taylor, & Kanarek, 2014), which in turn decreases couples’ capacity for conflict resolution (Forgas, 1994; Forgas & Tan, 2011; Joseph & Cromer, 2004).

Our affect infusion results suggest that a third mechanism links stress to marital deterioration: affect infusion itself. Clearly, stress instigates negative moods, and negative moods fundamentally alter the way people think about their relationships. Although relationship researchers have considered the effects of stress on relationship cognitions (e.g. Neff & Karney, 2004), they have yet to consider how the postulates of the AIM—and now the DAID model—explain such cognitions. For example, the AIM posits that evaluations emerging from “direct access” strategies—i.e. crystallized evaluations that are stored in memory—are immune to affect infusion. This suggests that couples with a persistent and easily accessible positivity bias about their partners (Martz, Verette, Arrriaga, Slovik, Cox, & Rusbult, 1998) might be those couples who are buffered from the negative effects of stress. Our work implies that theoretical postulates such as this could be applied to research on close relationships and stress.

Caveats and Further Considerations

As with all empirical studies, it is important to consider not only that these effects emerged, but also their magnitude. The largest effect was for affect infusion for examinees and their negative relationship evaluations. At the start of the diary, a one-unit difference in mood was associated with 1/3 of a unit difference in relationship evaluations. Over the course of the study, this effect decreased to be 1/6 of a unit immediately before the stressor. While these effects are modest, they are meaningful because they could inform intervention studies that protect couples from stress. Currently, interventions either target couples by giving them skills to build interpersonal and intraindividual strength (Bodenmann & Shantinath, 2004) or by targeting stress at the source (Lavner & Clark, 2017; Lavner, Karney, & Bradbury, 2015). Our results suggest that another potential way to mitigate the negative effects of stress would be to make salient the reason for each partner’s bad mood (for a deep discussion of the importance of small-medium effect sizes, see Greenwald, Banaji, & Nosek, 2015; Prentice & Miller, 1992).

At the same time, we tested the DAID model in a particular context, both in terms of relationship type and in terms of mood induction. How might relationship type influence main effects of affect infusion/diffusion? In terms of affect infusion, the original AIM asserts that the primary determinant of affect infusion is the cognitive processing strategy (Forgas, 1995), and specific predictions about the effect of relationship type were vague at best.⁵ Fortunately, other research implies a clearer hypothesis; namely that affect infusion should be strongest in close, communal relationships as this is where emotions occur most frequently (Bercheid & Ammazzalorso, 22001) and exert their strongest influence (Marroquín & Nolen-Hoeksema, 2015). Since we only collected data for close, communal relationship partners, future studies can test whether affect infusion is moderated by relationship type. Close relationships are also those in which affect diffusion might be strongest, as emotions are most often disclosed and expressed in these types of relationships (Clark & Finkel, 2004; Rimé, 2009; Von Culin, Hirsch, & Clark, 2018), at least in part because partners feel a sense of communal responsibility for the other’s needs (Clark, Fitness, & Brissette, 2001).

In terms of the effect of attributional salience varying by relationship type, prior research suggests that motivation might matter. As discussed previously, people are particularly motivated to view members of certain relationships in a positive light, especially their close relationship partners and members of their ingroups (Cichocka, 2016; Song et al., 2019). It is therefore possible that attribution only moderates affect infusion/diffusion for relationships with persistent motivation for positivity. Indeed, the original AIM has been criticized for overlooking the key role of motivation in affective and cognitive processing (Isen, 2002), and newer research suggests that motivational styles do moderate important relationship processes such as the efficacy of invisible versus visible social support (Zee, Cavallo, Flores, Bolger, & Higgins, 2018). Future studies could similarly explore whether moderation of affect infusion/diffusion by attributional salience depends on motivations within the relationship.

Peculiarities of the mood induction might also matter, albeit more so for moderation of affect infusion/diffusion rather than the main effects. In our empirical study, we tested the effects of a stressor that mostly impacted one partner and that had a with a clear offset. Although these types of stressors are quite common (i.e. one partner undergoes a scheduled surgery, one partner loses a parent, one partner faces an impending job loss), it is relevant to consider whether this structure posed a boundary condition. In terms of main effects of affect infusion/diffusion, we think it unlikely that the induction type matters so long as the mood of one or both partners is affected. If anything, the naturalistic mood induction that we studied might have been relatively weak in that only one partner was severely impacted, and stressors that mutually influence both partners (i.e. an impending eviction for both) might exhibit even greater effects insofar as both partner’s moods are directly impacted. Nonetheless, the clear offset of the stressor may have implications for moderation of affect infusion/diffusion by attributional salience.

Specifically, attributions may be adaptive when partners know that the stressor will end and that their partners or themselves can “return to normal”. However, this attributional process may not function adaptively in a chronic stress context. In the case where the stressor is unlikely to end (i.e. working multiple jobs at once, being consistently unable to pay bills, experiencing protracted racial discrimination, etc.), it seems insufficient to suggest that couples could simply “appraise their way” out of their negative moods and subsequent negative evaluations of their relationships. Indeed, recent research on the neuropsychology of stress suggests that individuals experiencing chronic stress show less activation in brain regions associated with emotion reappraisal (Perchtold et al., 2018), which implies that reappraisal for chronic stress is impaired at the neurological level. Moreover, classic research on learned helplessness postulates that individuals who try to exert control over an uncontrollable environment actually experience reversals in efficacy and in their sense of control (Abramson, Seligman, & Teasdale, 1978; Seligman, 1972). As such, continually attributing negative mood to a persistent exogenous stressor might actually make people more stressed, which would have negative implications for their relationship evaluations. We therefore hesitate to imply affect infusion/diffusion would always be moderated by attributional salience, as doing so could trivialize the gravity of chronically stressful contexts.

Finally, in laying out the theoretical DAID model, we hope that future researchers consider how these effects might vary depending on how each of the constructs is operationalized. Future researchers could test how different types of emotions propagate through the system, whether these effects are cyclical, and precisely what kinds of social evaluations are implicated. For example, one partner’s depression could correspond with the other partner’s anxiousness, with each mood having unique effects on evaluative outcomes. In this case, would covariation between depression and anxiousness imply unique affect infusion and diffusion effects? Moreover, affect does not only influence social evaluations, but social evaluations also influence affect (Forgas, 2012). Do the negative social evaluations emerging from dyadic moods, in turn, lead to further affect covariation? While we examined negative mood and relationship evaluations, future studies could operationalize affect and interpersonal evaluations in different terms in order to explore these possibilities.

Finally, it is important to acknowledge that our sample is not representative of the general population but instead typifies a privileged class with access to higher education willing to participate in a daily diary study. Extrapolating these findings to all couples might therefore be inappropriate, especially since researchers have recently begun to expose fundamental processes in close relationships that are moderated by participants’ socioeconomic status, for example the finding that explicitly communicating relationship demands is only effective for couples with the social, educational, and economic means to address such requests (Ross, Karney, Nguyen, & Bradbury, 2019). At the same time, we find it unlikely that the effect of affect on interpersonal evaluations depends on socio-economic context, although future researchers could think more deeply about this implication.

Conclusion

This paper introduced the Dyadic Affect Infusion/Diffusion (DAID) model to investigate the role of affect on interpersonal evaluations at the dyadic level. The DAID model makes several theoretical and empirical contributions to subfields of psychology as varied as emotion, cognition, close relationships, and stress. Mood appears to play a critical role in people’s evaluations of others and their relationships with them, and these affect-infused judgments are modulated by the salience of a reason for those moods. Taken together, this research sheds light on the dynamic and dyadic nature of affective influences on interpersonal evaluations.