The affective dynamics of reassurance-seeking in real-time interactions

Major depressive disorder (MDD) affects nearly a fifth of the U.S. adult population in their lifetimes (Kessler, Petukhova, Sampson, Zaslavsky, & Wittchen, 2012). While depression is associated with significant functional impairment generally (Kessler, 2012; Kessler & Bromet, 2013), it exerts some of its most profound negative effects on interpersonal functioning. For example, depression is associated with smaller social networks (Visentini, Cassidy, Bird, & Priebe, 2018), social environments characterized by negativity, conflict, and rejection (Davila, Stroud, & Starr, 2009), and dissatisfaction in romantic relationships (Whisman, 2001). Moreover, stressors in the interpersonal domain (e.g., major arguments with close others) are uniquely associated with the onset and exacerbation of depression symptoms (Liu & Alloy, 2010). Consequently, elucidating behavioral mechanisms that drive stress and conflict in the romantic relationships of depressed individuals is critical for understanding how MDD develops and worsens.

Several interpersonal theories of depression posit that excessive reassurance-seeking (ERS), the behavioral tendency to repeatedly and relentlessly seek others’ assurances that the self is worthy, likable, and cared for, is a central mechanism in the generation of stress in the interpersonal relationships of depressed individuals (Coyne, 1976; Evraire & Dozois, 2011; Gillett & Mazza, 2018; Joiner & Metalsky, 2001). The overarching tenets of these theories are (a) ERS occurs as an attempt to gain assurances of worth, reduce negative affect, and perhaps maintain the relationship, but (b) ERS paradoxically induces negative affect in relationship partners. While depressed and nondepressed individuals may engage in reassurance-seeking, the generation of partner negative affect is only theorized to occur when depressed individuals (but not nondepressed individuals) seek reassurance. Close others may initially provide reassurance, but depressed individuals often doubt the sincerity, leading to increased discomfort and need to seek yet more reassurance. This creates a cycle where reassurance is sought, received, and doubted with more frequency or force, producing greater negative affect in the partner (i.e., negative affect contagion; Hammen, 1991; Joiner, 1994). Through this mechanism of negative affect contagion, ERS subsequently predicts (c) more frequent interpersonal stressors, ultimately leading to rejection by the relationship partner. The resulting relationship erosion or rejection (d) further exacerbates depression symptoms in the depressed individual. In sum, these models propose two stages of negative outcomes where reassurance-seeking has long-term negative outcomes (i.e., rejection; increased depression) mediated by short-term (i.e., negative affect contagion) negative outcomes among individuals with high depression symptoms. The amount of rejection depressed reassurance-seekers experience is predicted by their partner’s increase in negative affect.

The negative long-term consequences of trait ERS, tenets (c) and (d) above, are robustly supported by research. Cross-sectional and longitudinal studies suggest that trait ERS is associated with more severe depression symptoms (Evraire & Dozois, 2014; Haeffel, Voelz, & Joiner, 2007; Hames, Hagan, & Joiner, 2013; Starr, 2015) and higher levels of interpersonal stress, negative interpersonal life events, and rejection for the depressed individual (Birgenheir, Pepper, & Johns, 2010; Evraire & Dozois, 2011; Shaver, Schachner, & Mikulincer, 2005; Starr & Davila, 2008; Stewart & Harkness, 2015). Similarly, the association between self-reported depression symptoms and concurrent self-reported trait ERS is well-documented in clinical and nonclinical samples (Evraire & Dozois, 2014; Fowler & Gasiorek, 2017; Joiner & Metalsky, 2001; Starr & Davila, 2008).

This body of literature has established the relation between reassurance-seeking and the hypothesized long-term negative outcomes across several interpersonal domains (e.g., friendships, family relationships, and romantic relationships); however, the strongest and most consistent support emerges in the context of long-term, heterosexual romantic relationships (Marcus & Nardone, 1992; Segrin & Dillard, 1992; Starr & Davila, 2008). Specifically, studies of romantic relationships have found that trait ERS predicts lower partner-rated relationship satisfaction and worse partner appraisals in romantic relationships (Benazon, 2000; Fowler & Gasiorek, 2017; Katz & Beach, 1997; Katz, Beach, & Joiner, 1998, 1999; Stewart & Harkness, 2017).

In contrast, only two studies have tested the hypothesis that ERS is a mechanism of negative affect contagion. First, Joiner (1994) found roommates of depressed individuals exhibited increased depression over 3 weeks, especially if the depressed individual endorsed high trait ERS. Similarly, Katz, Beach, and Joiner (1999) found a positive, cross-sectional association between a depressed individual’s and their romantic partner’s depressive symptoms, which was stronger for depressed individuals who endorsed high trait ERS. Taken together, these findings suggest that reassurance-seeking may exacerbate negative affect contagion. However, these studies’ methodologies are not consistent with Coyne’s (1976) view of negative affect contagion as a dynamic process, such that reassurance-seeking induces negative affect in the partner in the moment.

Methodological limitations of the extant literature

The literature reviewed above provides empirical support for the relations between self-reported trait ERS, depression, and interpersonal stress/rejection as well as limited support for the negative affect contagion hypothesis. Nevertheless, the field as a whole is limited by several interrelated methodological considerations that constrain conclusions about whether reassurance-seeking as a behavior induces negative affect in relationship partners in real time (Coyne, 1976; Evraire & Dozois, 2011; Gillett & Mazza, 2018; Joiner & Metalsky, 2001). These limitations include (1) lack of differentiation between reassurance-seeking and ERS, (2) lack of behavioral measures of ERS, (3) lack of theoretically coherent measurement timeframes, and (4) suboptimal operationalization of constructs.

First, throughout the literature reviewed here it is unclear what distinguishes reassurance-seeking from ERS. The term “excessive” may suggest elevated frequency or intensity; however, it is theorized that reassurance-seeking, even if engaged in frequently, is tolerable in the absence of depression because it does not come with the desperation proposed to be interpersonally toxic (Evraire & Dozois, 2011; Hames et al., 2013). In fact, average estimates of ERS are quite low across the literature, suggesting that frequency is irrelevant to the documented negative outcomes of ERS (Evraire & Dozios, 2011). It may be that reassurance-seeking is only perceived as excessive when one has particular depressogenic schemata (i.e., negative core beliefs about the self, world, and future) and/or in the context of other depressive behaviors (e.g., negative feedback seeking; see Joiner, Alfano, & Metalsky, 1993; Swann, Griffin, Predmore, & Gaines, 1987). ERS may not appear objectively distinct from normative reassurance-seeking; instead, it may be the context of the behavior that defines it as excessive.

This contextual information is not captured in the extant literature, partly because studies almost exclusively assess ERS with the 4-item ERS scale of the Depressive Interpersonal Relationships Inventory (DIRI-ERS; Joiner, Alfano, & Metalsky, 1992). The DIRI-ERS assesses the frequency of reassurance-seeking behaviors via self-report, and therefore, assesses one’s perception of themselves as a reassurance-seeker. Self-presentation biases may lead to inflated or attenuated reports of behavior (Robins, Fraley, & Krueger, 2009) and do not allow for measurement of discrete behavioral occurrences of reassurance-seeking or affect in real time. Thus, it remains unclear whether observable, moment-to-moment reassurance-seeking behaviors occurring in romantic partner dyads induce negative affect in the partner.

Relatedly, the studies that measure ERS and its proposed negative sequalae typically assess these factors concurrently, reducing the ability to analyze relations over time. The two studies analyzing the negative affect contagion hypothesis test these relationships cross-sectionally (Katz et al., 1999) and over several weeks (Joiner, 1994), but these durations of measurement do not allow for analysis of the real-time antecedents and consequences of ERS in romantic partner interactions. It is unclear from the literature what exactly prompts reassurance-seeking behavior. It may be a general depressed mood, a discrete negative emotion, a general cognitive-behavioral set, or a discrete thought that prompts reassurance-seeking behavior, and theories do not consider the antecedent to reassurance-seeking as a necessary component in the generation of negative affect in the partner (Coyne, 1976; Evraire & Dozois, 2011; Gillett & Mazza, 2018; Joiner & Metalsky, 2001). Instead, the negative affect contagion hypothesis states that reassurance-seeking predicts negative affect in the partner in the context of high depression symptoms, generally. To properly test this hypothesis, it is necessary to measure the target’s (defined here as the depressed individual) reassurance-seeking behavior, and the partner’s (defined here as the romantic partner of the target) affect immediately following a target’s reassurance-seeking. Furthermore, the theory states that this association should only occur among depressed reassurance seekers (i.e., depression as a moderator).

Finally, negative affect is inconsistently measured and operationalized across the extant literature. A meta-analysis investigating the phenomena of negative affect contagion (irrespective of reassurance-seeking) demonstrated that studies have assessed negative affect as partner’s depression, anxiety, and hostility (see Joiner & Katz, 1999). Similarly, the two studies investigating the effect of reassurance-seeking on partner’s affect measured depression, anxiety, and general positive and negative affect (Joiner, 1994; Katz et al., 1999). Coyne (1976) viewed negative affect contagion as a dynamic process, such that reassurance-seeking induces negative affect in the partner in the moment but is not specific about the type of negative affect that reassurance-seeking may induce in dynamic interactions. One study aimed to assess this relation on a dynamic time scale. Lemay and Cannon (2012) experimentally manipulated ERS by leading members of romantic couples to believe their dysphoric partner asked them to respond to reassurance-seeking questions and found romantic partners responded to the ERS manipulation with heightened frustration and reduced acceptance of dysphoric partners (Lemay & Cannon, 2012). This suggests that partners may react negatively to ERS; however, the contagion hypothesis remains untested because the ERS did not occur naturally due to the dysphoric partner’s negative affect. In fact, it did not actually occur at all. Thus, the contagion hypothesis has never been tested at the appropriate time scale or in a dynamic real-time interaction as Coyne (1976) proposed. However, several behavioral investigations have provided preliminary insights into the effects of behavioral reassurance-seeking in the context of depression.

Behavioral investigations of Coyne’s (1976) model

Behavioral investigations offer insight on moment-to-moment targets for adjusting maladaptive behaviors in dyadic relationships. Interpersonal theories state that it is the reassurance-seeking behavior, not perceptions, that ultimately contribute to relationship distress, because negative affect contagion ultimately contributes to interpersonal rejection. Therefore, it is essential to parse out the effect of behavior from the effect of perceptions captured by self-reports. Identifying maladaptive interpersonal behaviors, their consequences and antecedents, may help to inform interventions for decreasing relationship distress, including self-help resources and clinical interventions (e.g., Fischer, Baucom, & Cohen, 2016). Moreover, understanding these behaviors in the context of depression can greatly enhance the ability of clinicians to address these behaviors directly (e.g., de Mello, de Jesus Mari, Bacaltchuk, Verdeli, & Neugabauer, 2005) to improve interpersonal functioning and reduce depression symptoms.

Joiner and Metalsky (2001) were the first to attempt to behaviorally code reassurance-seeking during interactions of college roommates. Their analyses found that the DIRI-ERS scale was moderately associated with observer-rated reassurance-seeking behavior, suggesting that ERS may be observable in real-time interactions. Further research has aimed to establish that depressed individuals engage in observable reassurance-seeking during interactions with their romantic partners. Knobloch, Knobloch-Fedders, and Durbin (2011) engaged romantic couple dyads in unstructured conversation coded for reassurance-seeking behaviors. Reassurance-seeking was defined as requesting or inviting praise, and codes were based on verbal and nonverbal cues. Reassurance-seeking was rated on a Likert-type scale from not at all (0) to extremely (6) for 2-min periods. Results indicated that both partners’ self-reported depression symptoms were positively associated with the target’s behavioral reassurance-seeking, providing partial support for the contagion hypothesis. Additionally, across three studies in which romantic partners discussed personal information, Girme, Molloy, and Overall (2016) coded behavioral reassurance-seeking, which they defined as statements that sought verification of one’s self-worth or the partner’s commitment. Codes were based on frequency, quality, and duration of statements when rating the overall degree of reassurance-seeking during the conversation. The authors found no relation between behavioral reassurance-seeking and target’s depression symptoms or partner-rated relationship quality but found a small, positive association between behavioral reassurance-seeking and the DIRI-ERS.

Finally, Stewart and Harkness (2017) employed a laboratory paradigm loosely based on the work of Joiner and Metalsky (2001) with a sample of romantic couples, coded women’s reassurance-seeking statements based on the criteria that the statements were related to a personal quality of the target, aimed at garnering positive feedback from the partner, and phrased in a way that limited potential partner responses. This study found that behavioral reassurance-seeking was associated with lower partner-reported relationship quality for women high in depression, partially supporting interpersonal theories. The mixed support for Coyne’s (1976) model in studies to date, and the relative dearth of research that behaviorally codes reassurance-seeking, precludes conclusions about the effect of behavioral reassurance-seeking on partners’ affect in real time. Furthermore, these studies still do not address whether reassurance-seeking behavior exhibited by depressed individuals evokes negative affect in their romantic partners because they do not assess the immediate effects of behavioral ERS relevant to interpersonal models of depression.

Current research

The present study extends previous research by providing the first investigation of the negative affect contagion hypothesis of Coyne’s (1976) interpersonal theory (i.e., that reassurance-seeking predicts negative affect in the partner only among depressed individuals). In this study, we also investigate whether the target’s discrete negative affect predicts their reassurance-seeking behavior. The present study draws its data from the same larger study on interpersonal processes in depression as the work of Stewart and Harkness (2017). The main overlap between the data analyzed is the coding of reassurance-seeking statements. Stewart and Harkness (2017) evaluated the associations between trait measures of ERS, depression, core beliefs, and partner-rated relationship quality with the frequency of behavioral reassurance-seeking to assess the long-term tenets of Coyne’s (1976) model (i.e., that ERS eventually leads to rejection, operationalized as lower partner satisfaction). The present study extends their findings by investigating real-time dyadic consequences of behavioral reassurance-seeking to test Coyne’s (1976) dynamic negative affect contagion hypothesis. Thus, the previous study investigated the end results of Coyne’s (1976) model, while the current investigates the means to that end. Additionally, to investigate these relations in real time, the present study used a micro-longitudinal design and examined the timing of reassurance-seeking during the task, as opposed to using a frequency count.

Specifically, we examine reassurance-seeking behaviors and affect in the context of a 5-min discussion task featuring females (target participant) and males (partners) in heterosexual romantic relationships. We incorporate methods that extend previous studies that have employed behavioral reassurance-seeking tasks (BRSTs) in several important ways. First, we code reassurance-seeking and both partners’ affect for ten 30-s epochs. This duration of assessment is shorter than that employed by other behavioral paradigms assessing reassurance-seeking (Girme, Molloy, & Overall, 2016; Joiner & Metalsky, 2001; Knobloch, Knobloch-Fedders, & Durbin, 2011) and allows for sophisticated and nuanced temporal analyses of relevant relationships.

Additionally, we operationalized negative affect using a validated paradigm into three specific types: anger/contempt, anxiety, and hurt. These codes are informed by theories that suggest reassurance-seeking may be produced by and produce frustration, anxiety, and emotional pain or sadness (e.g., hurt; Coyne, 1976; Evraire & Dozois, 2011; Gillett & Mazza, 2018; Joiner & Metalsky, 2001) and consistent with previous studies that have assessed a variety of negative affect (Joiner, 1994; Katz et al., 1999). Assessing three specific and varied types of negative affect allows more nuanced analysis of the precusors and consequences of reassurance-seeking.

Second, we create an experimental situation that is ambiguous and mildly socially threatening, instead of unstructured conversation. Third, we utilize verbal content as our basis for coding ERS, which avoids conflation with nonverbal affective behaviors, and we identify discrete instances as opposed to simply coding the frequency and duration of statements. Fourth, we use ERS coding criteria more specific to Coyne’s (1976) model in process (e.g., closed-ended questions) and content (e.g., personally relevant and seeking positive feedback). Finally, we code real-time affect, which has not previously been done. This allows an investigation of the relation between verbal reassurance-seeking content and nonverbal affective signals.

Consistent with Coyne’s (1976) negative affect contagion theory, the study’s primary hypotheses are that the target’s reassurance-seeking will predict subsequent increases in the male partner’s negative affect (e.g., anger/contempt, anxiety, and hurt) among individuals with high, but not low, depression symptoms. Specifically, female reassurance-seeking will predict increased male partner negative affect, among females with high depression symptoms, but not among those will low depression symptoms. We also test the secondary hypothesis that the female partner’s (the target participant) negative affect (e.g., anger/contempt, anxiety, and hurt) will predict subsequent reassurance-seeking behavior and that this relationship will be stronger among individuals with high, but not low, depression symptoms. Due to this being the first study of its kind, and due to theory and empirical research suggesting reassurance-seeking may be associated with various types of negative affect, our hypotheses do not vary across types of negative affect measured here. Finally, we present exploratory analyses of the relationships between female reassurance-seeking and male positive affect; and female positive affect and reassurance-seeking. Based on Coyne’s (1976) model, we expect that reassurance-seeking may not only produce partners’ negative affect but also reduce partners’ positive affect.

Method

Participants

The present study used data collected as part of a larger study on interpersonal processes in depression (Stewart & Harkness, 2015, 2017). Recruitment focused on female undergraduate students in heterosexual romantic relationships lasting 1 month or more at a Canadian university. Research suggests that in Canada women are nearly twice as likely to experience major depression compared to men (Patten et al., 2006, 2015) and that the relation between ERS and depression may be stronger for women (Starr & Davila, 2008). Therefore, women were chosen as the target participants in this nonclinical sample to increase the likelihood of recruiting individuals with a range of depression symptoms and the related likelihood of observing reassurance-seeking behaviors. Additionally, recruitment oversampled for women with high depression symptoms. The resulting sample consisted of 121 women (M _age = 18.63 [SD = 1.18], range = 17–23; 74.4% White, 14.9% Asian, 0.8% Hispanic/Latina, 9.9% Other) and their male partners (M _age = 19.25 [SD = 1.91], range = 17–26; 76% White, 9.1% Asian, 2.5% Black, 1.7% Hispanic/Latino, 10.7% Other). A majority of couples (59.3%) had been dating for at least 1 year (M _length = 16.27 months [SD = 10.84]). At the time of participation, 5.1% of couples were cohabiting and 22.9% were living in different cities.

Measures

Beck Depression Inventory–II (BDI-II)

The BDI-II is a 21-item self-report questionnaire that measures depression symptom severity in the past 2 weeks. Items are rated on a 4-point Likert-type scale from 0 to 3 (Beck, Steer, & Brown, 1996). Total BDI-II scores range from 0 to 63 with higher scores representing more severe depression. The BDI-II has good psychometric properties (Beck et al., 1996). While both the male and female partners completed the BDI-II, only the female partner’s scores were analyzed (M = 12.9, SD = 10.0, range = 0–46).

Behavioral Reassurance-seeking Task (BRST)

In the task, participants were told they would be assisting in the validation of a personality measure for undergraduate students (Stewart & Harkness, 2017). Both partners separately completed a 24-item sham personality questionnaire (about 5 min). Then they were escorted into a room, reunited, and asked to sit quietly while the questionnaires were scored, and a personality profile was generated for each partner. After 5 min, the experimenter returned with the female participant’s sham personality profile (identical for all participants). The sham profile included female participants’ scores on eight face-valid personality descriptors (e.g., Edgy, Active, Moody; see Supplemental Online Material). Participants discussed the results of the female’s profile for 5 min and were told to focus on getting each other’s opinions about how well the results fit the female’s personality. Analyses of reassurance-seeking focused on audio-recorded verbal interaction during the 5-min discussion period that started when the experimenter left the room after delivering the profile. Finally, the experimenter returned and debriefed the couple regarding the deception used in the task. The full coding manual is available in Supplemental Online Material.

Coding of audiotaped reassurance-seeking

The statements of the female partner were transcribed and evaluated for the presence or absence of reassurance-seeking based on three necessary criteria. In order to be considered reassurance-seeking (coded “1”), statements must have (a) sought personal information, (b) sought positive feedback, and (c) allowed for limited potential available responses (e.g., yes or no responses). For more details on the coding procedure, see Supplemental Online Material.

Each coded statement was associated with a time stamp that indicated when in the 5-min discussion period it occurred. To temporally align these codes with behaviorally coded affect (see below), we created a variable that indicated whether reassurance-seeking occurred (or not) during each of the ten 30-s epochs of the 5-min discussion period. For example, an instance of reassurance-seeking occurring 22 s after the audio recording began would be assigned a value of “1” (i.e., present) in the first epoch. We used this categorical coding approach because a participant indicating more than one instance of reassurance-seeking in an epoch was rare (6.2%). In total, there were 252 (20.8%) epochs coded as including an instance of female partner reassurance-seeking; and the percentage of participants who engaged in reassurance-seeking in a given epoch ranged from 9.0% (epoch 10) to 34.7% (epoch 2). To determine inter-rater reliability for coding reassurance-seeking behaviors, 65 (53.7%) audio files were coded by two raters. Raters agreed on 91% of the transcribed statements (κ = .73; see Stewart & Harkness, 2017 for further details).

Behavioral Affective Rating Scale (BARS)

The BARS is a rating system designed for analyzing romantic couples’ interactions by assessing affect based on the magnitude of specific behaviors (Johnson, 2002). We used the BARS to code positive (affection and humor) and negative (anxiety, defensiveness, aggressiveness, scorn, frustration, and hurt) affect on a scale from absent (0) to extreme (4) for each 30-s epoch. The BARS relies entirely on behaviors—including body language, facial expressions, and tone of voice—and ignores verbal content for deriving affect codes. These codes were assigned to the female and male partners independently for each of the ten 30-s epochs beginning when the experimenter left the room.

Previous factor analytic research indicates that the eight affect codes can be grouped into four factors: Anger/contempt, Anxiety, Hurt, and Humor/affection (Johnson, 2002). The Anger/contempt score is equal to the sum of the defensiveness, aggressiveness, scorn, and frustration scores, while the Humor/affection score is equal to the sum of the humor and affection scores. The Anxiety and Hurt scores are equal to their individual codes, respectively. Videos were coded independently by a team of three raters, and reliability was assessed by having the whole rating team code a subset of 10 videos (8.26%) for male and female affect. Intra-class correlations (ICCs) for absolute agreement were as follows for male affect codes: 0.65 for Anger/contempt, 0.66 for Anxiety, 1.00 for Hurt, and 0.90 for Humor/affection, and as follows for female affect codes: 0.50 for Anger/contempt, 0.71 for Anxiety, 1.00 for Hurt, and 0.78 for Humor/affection. These values suggest moderate agreement for all codes except female and male Humor/affection, which indicate good to excellent agreement (Koo & Li, 2016).

Consistent with recruiting self-selected couples, participants displayed positive affect much more frequently than negative affect. On average, across all epochs, females displayed more Humor/affection (M = 2.36 [SD = 1.91]), than Anxiety (M = 0.71 [SD = 0.74]), Anger/contempt (M = 0.17 [SD = 0.47]), or Hurt (M = 0.02 [SD = 0.13]), and males displayed more Humor/affection (M = 2.48 [SD = 1.91]), than Anxiety (M = 0.51 [SD = 0.65]), Anger/contempt (M = 0.07 [SD = 0.30]), or Hurt (M = 0.00 [SD = 0.00]). Specifically, Humor/affection occurred (code of 1 or higher) during 74.1% of all epochs for females and 75.8% for males. In contrast, Anxiety and Anger/contempt only occurred during 53.8 and 13.1% of all epochs for females and 41.8 and 5.5% of all epochs for males. Hurt was rarely expressed by females across all epochs (1.6%) and was not expressed by any male participants in any epoch. Thus, analyses focused on Anxiety, Anger/contempt, and Humor/affection only.

Procedure

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. Ethics approval was obtained at the time of the original study through approval by the General Research Ethics Board at Queen’s University, and informed consent was obtained from all couples. As part of a 2-hr laboratory session, couples completed questionnaires in separate rooms, including the sham personality profile, after which they were reunited for the BRST protocol.

Data analyses

To test associations between reassurance-seeking and partners’ affect, multilevel regression models were evaluated using the Hierchical Linear and Non-Linear Modeling Software Program (HLM version 7, Scientific Software International, 2016). Logistic multilevel regression models were evaluated when reassurance-seeking (coded as 0 for absent and 1 for present) was the outcome variable. Full-maximum likelihood estimation was used for models with continuous outcomes (i.e., BARS ratings). Reassurance-seeking codes were assigned for each female participant across ten 30-s epochs for the entirety of the 5-min BRST. Similarly, BARS codes for female and male participants were assigned to each 30-s epoch. Lagged variables were created for all affect and reassurance-seeking codes. First, we conducted preliminary analyses to investigate whether male and female affect were correlated within the interaction and whether male affect prospectively predicted changes in female affect while controlling for previous female affect, and vice versa. Then, to evaluate temporal dynamics between reassurance-seeking and male behavioral affect, and female behavioral affect and reassurance-seeking within the 5-min discussion period, respectively, multilevel lagged regression was utilized to analyze the data. The Level-1 (within participants) components consisted of the repeated female and male BARS codes, and female reassurance-seeking codes, which were nested in participants (Level-2). These analyses test the associations between female reassurance-seeking at time t and male BARS (Anger/contempt, Anxiety, and Humor/affection) at time t + 1, and female BARS (Anger/contempt, Anxiety, and Humor/affection) at time t and female reassurance-seeking at time t + 1, respectively. Finally, to test whether these relationships are moderated by female depression symptoms, female BDI-II scores were entered as a Level-2 moderator of all Level-1 relationships. We reported partial regression coefficients (pr) as effect size indicators for the multilevel regression coefficients. Kirk (1996) suggested .10, .24, and .37 for small, medium, and large effect sizes, respectively.

Results

Preliminary analyses

Neither female and male Anger/contempt (b = −.003, SE = .05, t[114] = 0.07, p = .946, pr = .01), Anxiety (b = .06, SE = .03, t[114] = 1.64, p = .103, pr = .15), nor Humor/affection (b = .06, SE = .04, t[114] = 1.63, p = .106, pr = .15) were correlated within the dyad. Regarding prospective relationships, male Anger/contempt did not predict subsequent female Anger/contempt over time, when controlling for previous female Anger/contempt (b = .06, SE = .08, t[114] = 0.77, p = .445, pr = .07); however, female Anger/contempt predicted subsequent male Anger/contempt, when controlling for previous male Anger/contempt (b = −.02, SE = .01, t[114] = 2.12, p = .036, pr = .19). Male Anxiety predicted subsequent female Anxiety, when controlling for previous female Anxiety (b = .07, SE = .03, t[114] = 2.11, p = .037, pr = .19), and female Anxiety predicted subsequent male Anxiety, when controlling for previous male Anxiety (b = .07, SE = .02, t[114] = 2.93, p = .004, pr = .26). Finally, male Humor/affection did not predict subsequent female Humor/affection, when controlling for previous female Humor/affection (b = −.01, SE = .03, t[114] = .43, p = .671, pr = .04); however, female Humor/affection significantly predicted subsequent male Humor/affection, when controlling for previous male Humor/affection (b = .08, SE = .04, t[114] = 2.13, p = .035, pr = .20). In sum, the affect of the female target impacted the subsequent affect of the male partner, but male partner’s affect seemingly did not influence female target’s affect with the exception of female and male Anxiety reciprocally and positively predicting each other.

Female reassurance-seeking predicting male negative affect

The association between female reassurance-seeking and subsequent male Anxiety approached statistical significance (b = .08, SE = .04, t[114] = 1.92, p = .057, pr = .18). However, female reassurance-seeking did not significantly predict subsequent male Anger/contempt (b = −.01, SE = .02, t[114] = .86, p = .391, pr = .08) or male Humor/affection (b = .05, SE = .11, t[114] = .45, p = .657, pr = .04).

The negative affect contagion hypothesis: Depression as a moderator

Multilevel lagged regressions with female reassurance-seeking as a categorical Level-1 predictor of male affect and female BDI-II scores as a Level-2 moderator indicated that depression did not moderate the association between female reassurance-seeking and male Anxiety (b = .0003, SE = .004, t[113] = .08, p = .934, pr = .01), male Anger/contempt (b = −.0004, SE = .002, t[113] = .24, p = .814, pr = .02), or male Humor/affection (b = .002, SE = .01, t[113] = .12, p = .906, pr = .01).

Female affect predicting reassurance-seeking

Results indicated that female Anxiety at one time was associated with subsequent reassurance-seeking in the next epoch (b = .26, SE = .12, t[114] = 2.15, p = .034, pr = .20). However, female Anger/contempt did not significantly predict subsequent reassurance-seeking (b = −.03, SE = .19, t[114] = 0.17, p = .869, pr = .02). Female Humor/affection also evidenced no relation with subsequent reassurance-seeking (b = .06, SE = .05, t[114] = 1.15, p = .251, pr = .11).

A multilevel lagged regression with female Anxiety as a Level-1 predictor of female categorical reassurance-seeking and female BDI-II scores as Level-2 moderator indicated that BDI mitigated the positive relationship between female Anxiety and female reassurance-seeking (b = −.03, SE = .01, t[113] = 3.17, p = .002, pr = .28). The nature of this BDI × Female Anxiety interaction was such that a weaker association between female Anxiety and female reassurance-emerged for females who endorsed more severe depression symptoms relative to females who endorsed less severe depression symptoms. Depression did not significantly moderate the association between female Anger/contempt and female reassurance-seeking (b = .002, SE = .002, t[113] = 1.21, p = .228, pr = .11) nor did it moderate the association between female Humor/affection and female reassurance-seeking (b = .00003, SE = .001, t[113] = .04, p = .967, pr = .004).

Discussion

Studying interpersonal processes within the context of depression is essential to understanding the development and worsening of this condition. We examined negative affect contagion behaviorally, which affords a nuanced understanding of the potentially deleterious effects of depression-driven behaviors on interpersonal relationships. We found that female anxiety predicted subsequent reassurance-seeking, and there was a trend for female reassurance-seeking to predict male partners’ subsequent anxiety. Further, depression symptoms moderated the relation between female anxiety and reassurance-seeking. Surprisingly, this association was stronger among women with less severe current depression symptoms. These findings offer novel insights into the immediate antecedents and consequences of reassurance-seeking in real-time interactions within and beyond the context of depression.

Primarily, these findings may help to elucidate the difference between routine, effective reassurance-seeking behaviors, and ERS. The behavioral phenomenon captured in this study may better represent reassurance-seeking as opposed to ERS in part because the mean level of depression symptoms was in the minimal range, and female negative affect was coded infrequently. Reassurance-seeking may only take on an excessive tone in the presence of depression and over a long period of time. Thus, in our sample of self-selected, non-distressed couples with female partners who endorsed relatively low levels of depression, it is understandable that female reassurance-seeking in a discrete, mildly threatening interaction did not result in negative affect in male partners. ERS may not be directly associated with relationship distress; instead, it may worsen conflict and increase the likelihood of rejection when reassurance seekers are in maladaptive relationships. It is critical that future research continues to differentiate between adaptive reassurance-seeking and ERS.

Additionally, these findings provide support for Evraire and Dozois’s (2011) and Gillet and Mazza’s (2018) revised theories of ERS. Evraire and Dozios (2011) posit that ERS can be elicited by an intolerance of uncertainty, the predisposition to negative emotional, behavioral, or cognitive reactions to ambiguous events. This intolerance may lead to increases in anxiety in response to threats to self-esteem, which results in ERS to reduce this distress as quickly as possible. Similarly, Gillet and Mazza’s (2018) integrative functional model of reassurance-seeking suggests that reassurance-seeking can dually function to preserve self-worth and reduce anxiety about interpersonal rejection or abandonment. As such, ERS in depression may not only be elicited by depressed affect along with doubts that the self is worthy, likable, and cared for as suggested by Coyne (1976) but also be elicited by anxiety about social threats. Additionally, ERS elicited by anxiety may similarly generate negative affect in relationship partners and long-term interpersonal relationship stress.

Anxiety as defined in this study included observable behaviors indicating nervousness, tension, and/or discomfort, such as an anxious tone of voice or fidgeting. This emotional and behavioral reaction may reflect intolerance of uncertainty, which is not surprising given that the BRST creates an ambiguous, and potentially socially threatening situation. Female partners may have felt uncertain if they believed their sham profiles did not necessarily align with their self-perceptions of their own personality or if they found the ambiguous feedback threatening. The feedback is somewhat unpredictable and uncontrollable in that females could not predict or edit the feedback or what their male partners perceived about the feedback. Intolerance of uncertainty was not measured here, but it is possible that discomfort with that uncertainty may have led women to ask their partner for positive feedback to immediately neutralize distress. Consequently, this request for positive feedback may have generated mild distress in male partners, as evidenced by a trend toward male partners exhibiting behavioral signs of anxiety in response to female reassurance-seeking.

Contrary to Coyne’s (1976) theory, women with more severe depression evidenced a weaker relationship between their anxiety and reassurance-seeking. It may be that ERS is characteristic of those with moderate, subclinical depression symptoms. Few women in our sample endorsed severe symptoms (approximately 9%). Although we did not compare between groups based on depression symptoms, previous research on ERS in clinical samples have found either small or nonsignificant effects of negative affect on ERS (Starr & Davila, 2008). Individuals with more severe depression symptoms may have decreased motivation to seek positive feedback or find positive feedback highly discordant with their negative self-concepts (Evraire & Dozois, 2011; Swann et al., 1987). Additionally, individuals with more severe depression symptoms may be more inclined to avoid negative information, and as such are less likely to seek feedback as they see the risk of it being negative as highly aversive.

The relations between female Anger/contempt and ERS and female ERS and male Anger/contempt were not significant even among individuals with high, depression symptoms. Anger/contempt here consisted of behavioral indicators of frustration, aggression, defensiveness, or scorn, which are the types of reactions hypothesized to be the culmination of negative affect contagion. There are several possible explanations for why these relations were not detected. First, it is possible that frustration arises over longer time frames with increased frequency of ERS or that it is actually a result of the partner doubting the sincerity of the reassurance given (Benazon, 2000). It may also be that ERS results in partner negative affect only when the partner feels responsible for managing the others’ insecurities (Lemay & Cannon, 2012), and this behavioral task did not elicit such a partner response.

Additionally, the study sample included a self-selected group of couples in generally happy relationships who endorsed low instances of anger and contempt during the interaction. Thus, negative affect approached the statistical floor, and there was limited variance in these scores. Couples in relationship distress may be more likely to evidence a range of affect and exhibit a significant relationship between ERS and negative affect, as has been documented in the literature (Benazon, 2000; Fowler & Gasiorek, 2017; Katz & Beach, 1997; Katz et al., 1998, 1999). Finally, the study task was not overtly a conflict task, and as such, high levels of tension were not introduced. A more socially threatening task may have elicited greater frustration. For example, Stewart and Harkness (2017) suggest using a task that manipulates feelings of interpersonal insecurity, such that participants are told their sham personality feedback means they will end up alone (Twenge, Baumeister, DeWall, Ciarocco, & Bartels, 2007).

Moreover, these findings provide insights into the theoretical basis of the negative affect contagion hypothesis and ERS more broadly. First, we found limited evidence that behavioral reassurance-seeking is prompted by negative affect. Coyne’s (1976) model was not specific on what prompts reassurance-seeking, and negative affect is unlikely to be the only thing that elicits ERS behavior. Additionally, negative affect might not elicit reassurance-seeking in all those who tend to reassurance-seek (for some, it might be stress, negative automatic thoughts, etc.). Future research should investigate the precipitants of reassurance-seeking to better understand its function(s).

Furthermore, we did not find strong evidence that behavioral reassurance-seeking induces negative affect in romantic partners in real time. There are some methodological limitations of the present study (described above) that could explain this lack of association. Alternatively, it is possible that ERS is not a behavioral phenomenon per se, but that the ERS measured by self-report questionnaires in previous research is actually better captured by cognitive constructs related to insecurity in relationships and fears of abandonment, including insecure or anxious attachment style, dependency, and negative cognitions about the self (Evraire & Dozios, 2011; Gillett & Mazza, 2018). In fact, research has found moderate positive correlations between anxious attachment and trait and daily ERS (Evraire & Dozois, 2014; Shaver et al., 2005; Stewart & Harkness, 2015). Future research should investigate whether these cognitive constructs may explain some of the negative outcomes attributed to ERS.

Finally, if it is the perception of reassurance-seeking and not the behavior that is associated with interpersonal stress, then it is possible that the function of reassurance-seeking behaviors varies across contexts. As others have suggested, reassurance-seeking may be an effective strategy for distress reduction and relationship maintenance especially when individuals do not feel depended on for their partners emotion regulation (Evraire & Dozois, 2011; Lemay & Cannon, 2012). One study used behaviorally coded reassurance-seeking in romantic partner dyads and found that greater reassurance-seeking repaired partner-reported lack of closeness for partners of highly attachment avoidant individuals. Additionally, in response to target reassurance-seeking, partners provided more reassuring support (Girme et al., 2016). Taken together, these findings may suggest that interventions to decrease relationship distress in the context of and beyond the context of depression may need to focus on targeting cognitive manifestations of ERS as opposed to changing behaviors.

Implications and limitations

Our findings have several important implications that contribute to the theoretical conceptualization and measurement of ERS. Our study extends previous literature in establishing that reassurance-seeking is an observable behavior. Prior studies have demonstrated a moderate, positive association between self- and other-reported reassurance-seeking (Evraire & Dozois, 2014; Joiner & Metalsky, 2001; Shaver et al., 2005) and have successfully behaviorally coded reassurance-seeking and found reassurance-seeking behavioral codes evidenced small to moderate positive associations with self-reported trait ERS (Girme et al., 2016; Joiner & Metalsky, 2001; Knobloch et al., 2011; Stewart & Harkness, 2017). This study built upon these findings by raising important theoretical and empirical considerations regarding the differentiation of the perception of ERS and the quantification of ERS behaviors for future research. Additionally, these findings support revised, integrative theories of ERS (Evraire & Dozois, 2011; Gillett & Mazza, 2018), which warrant further study. In particular, future research should focus on measuring attachment styles, and cognitive schemas that may be associated with ERS. The fact that anxiety predicted ERS here may suggest that ERS is a mechanism that arises in the context of the activation of anxious attachment or maladaptive schemas and functions to assuage fears of abandonment, rejection, or low self-worth (Evraire & Dozois, 2011; Girme et al., 2016).

Our findings should be interpreted in the context of the following limitations. First, our sample consisted of college students in heterosexual romantic relationships, which limits the generalizability of our findings to same-sex relationships, relationships across the life span, and other types of close relationships. Future research should investigate these mechanisms across different types of romantic and close relationships. In particular, theories of ERS have not yet been investigated with non-heterosexual romantic relationships. This is a particularly striking omission given that lesbian, gay, bisexual, transgender, or queer, and similarly identified people may experience higher rates of depression and other forms of psychological distress (Herek & Garnets, 2007; Woodford, Kulick, & Atteberry, 2015).

The social acceptability of ERS may differ as a function of age (e.g., more acceptable in children than adults) and relationship type (e.g., family relationships, friendships, and romantic relationships) (Starr & Davila, 2008). Additionally, our sample was racially and ethnically homogeneous, which may reduce generalizability across cultures. Furthermore, our sample was self-selected and consisted of individuals in non-distressed romantic relationships. Results may differ in romantic relationships in which one or both partners are distressed or dissatisfied. Second, we selected females as the target partner, which limits the generalizability of findings to the mechanism of ERS in males. Additionally, it was not possible to assess the impact of reassurance-seeking relative to the reassurance-seeking of the male partner. It may be that there are differing opinions on the acceptability of reassurance-seeking based on the relative behavior of both partners, such that reassurance-seeking is viewed as particularly negative if one partner reassurance-seeks frequently, while the other rarely does, but is tolerable if both partners frequently seek reassurance and view it as more routine.

Finally, the Hurt code was not analyzed here due to low endorsement. Hurt is defined as emotional pain and sadness and may have better represented instances of guilt and low self-worth, which are theorized to precipitate depressive ERS (Coyne, 1976; Joiner et al., 1992). Additionally, endorsement of the Anger/contempt code was also infrequent, which is likely related to the moderate ICC, such that infrequent disagreement would decrease the ICC in the light of few coded instances. This may have affected our ability to detect relationships with ERS.

Furthermore, the BRST (Stewart & Harkness, 2017) may not have elicited sufficient levels of social threat to create affective reactions leading to ERS, which is another direction for future research.

The present study is the first to examine the relationships between behavioral reassurance-seeking and the affect of romantic partner dyads in the context of a real-time interaction. This study addressed a long-standing empirical gap in interpersonal theories, as it is the first empirical test of the negative affect contagion hypothesis (Coyne, 1976; Joiner, 1994). The demonstrated relationships between female anxiety, ERS, and male anxiety primarily support revised interpersonal theories of reassurance-seeking (Evraire & Dozois, 2011; Gillett & Mazza, 2018) and help to elucidate the precipitants and antecedents of behavioral reassurance-seeking in romantic partner interactions.

Supplemental material

Supplemental Material, ReassuranceSeeking_CodingManual_FINAL - The affective dynamics of reassurance-seeking in real-time interactions

Supplemental Material, ReassuranceSeeking_CodingManual_FINAL for The affective dynamics of reassurance-seeking in real-time interactions by Kayla A. Lord, Kate L. Harkness, Michael K. Suvak and Jeremy G. Stewart in Journal of Social and Personal Relationships