The effect of social interaction quantity and quality on depressed mood and loneliness: A daily diary study

Interaction-Level Processes and the Interpersonal Process Model

Social interactions form the basic building blocks of our relationships, and thus, the multiple health benefits incurred by having strong, trusting relationships is a product of our repeated interactions with others over time. Delineating the psychosocial processes that occur at the level of the interaction, then, is crucial to developing a stronger scientific understanding of how relationships affect mental health and ultimately identifying mutable processes amenable to intervention. This is particularly important considering evidence that existing interventions for loneliness are only moderately effective (Masi et al., 2011) and that the public health burden of depression is growing despite increased utilization of psychiatric services (World Health Organization, 2008).

The Interpersonal Process Model (IPM; Reis & Shaver, 1988) is a theoretical model of the interaction-level psychosocial processes that contribute to the development of close, trusting relationships over time. In essence, the IPM suggests that close relationships are formed and maintained as a function of repeated interactions characterized by a reciprocal exchange of self-disclosure and responsiveness. According to the model, when one’s self-disclosures are met with responses that communicate understanding, validation, and care (i.e., responsiveness), interpersonal closeness is fostered. Research suggests that self-disclosures of emotion are stronger predictors of closeness than disclosures of facts and information (Beike et al., 2016), and that both perceived and enacted (i.e., objectively coded) responsiveness mediate the association between self-disclosure and interpersonal closeness (Canevello & Crocker, 2010; Laurenceau et al., 1998). The interpersonal process described by the IPM is broadly applicable across relationship types (e.g., romantic, friendships) and has been replicated over several decades of research across scientific disciplines (Kanter et al., 2020; Laurenceau et al., 2004).

The Current Study

The overarching aim of this research was to identify the components of daily social interactions that are associated with changes in depressed mood and loneliness. Gaining a deeper understanding of these components and how they unfold over time will ultimately lead to more precise prediction (e.g., of the effect of structural changes in our social relationships) and effective intervention strategies that target social relationships and mental health. To address this overarching aim, we (a) investigate the relative contribution of daily social interaction quantity and quality (defined per the IPM as vulnerable self-disclosure and perceived responsiveness) on changes in loneliness and depressed mood, (b) explore the effects of social interaction quantity and quality at the between-person and within-person levels of analysis, and (c) explore whether these effects interact to form a more complex picture of the role of our social interactions over time.

Participants and Procedure

Participants were adult residents of King County, Washington recruited through local news articles about the study, targeted advertisements on social media, and flyers distributed at grocery stores, restaurants, and other physical locations. Recruitment took place between March 14, 2020 and April 17, 2020, with the majority of enrollment (82%) occurring before March 25, 2020. Of the 543 participants who enrolled, 4 (1%) did not complete any surveys and 24 (4%) resided outside of King County and were excluded. The final sample comprised 515 individuals (M_age = 39.68 years, SD = 14.89, range = 18–81), the majority of whom were cisgender women (n = 360; 69.9%), white (n = 404; 78.4%), and employed at the time of enrollment (n = 381; 74.0%). Most participants were married (n = 219; 42.5%) or in a romantic relationship (n = 142; 27.6%) and lived with their family/romantic partner (n = 333; 64.7%) or friends/roommates (n = 65; 12.6%). Ninety participants (17.5%) self-reported a current or past diagnosis of Generalized Anxiety Disorder, Social Anxiety Disorder, or Obsessive-Compulsive Disorder, 32 (6.21%) reported a current or past diagnosis of Major Depressive Disorder, and 90 (17.5%) reported more than one of these diagnoses. See Table 1 for detailed participant demographic information.

OPEN IN VIEWER

Table 1.

Participant Demographic Information.

	N (%)
Gender
Cisgender man	116 (22.5%)
Cisgender woman	360 (69.9%)
Trans/non-binary/other gender	36 (6.99%)
Race/ethnicity
White	404 (78.4%)
Asian	51 (9.90%)
Black	8 (1.55%)
Hispanic	15 (2.91%)
Mixed race	35 (6.80%)
Income
Less than US$9999	62 (12.0%)
US$10,000–US$19,999	37 (7.2%)
US$20,000–US$39,999	69 (13.4%)
US$40,000–US$59,999	61 (11.8%)
US$60,000–US$79,000	91 (17.7%)
US$80,000–US$99,999	51 (9.90%)
More than US$100,000	136 (26.4%)
Relationship status
Partnered	361 (70.1%)
Single	152 (29.5%)
Living arrangement
With others	398 (77.3%)
Alone	104 (20.2%)
Parent
Yes (lives with children)	136 (26.4%)
Yes (does not live with children)	68 (13.2%)
No	309 (60.0%)

Participants were presented with a description of the study and provided informed consent to participate. All procedures were approved by the University of Washington’s institutional review board prior to data collection. Upon enrollment, participants were asked to complete daily surveys every evening for 75 consecutive days using their mobile phone. Each survey contained roughly 3-minutes’ worth of content, and items were presented randomly to minimize inter-item proximity effects (Weijters et al., 2009). Every evening at 7:30pm, participants received a text message that directed them to a secure online data collection platform. Participants were required to complete the survey before 5:00am the following morning, however exceptions were granted as necessary (e.g., for night shift workers) as long as participants completed the survey prior to the end of their day (i.e., falling asleep). Participants were provided with a daily text message reminder if they did not complete the survey by 10:30pm each evening, and additional reminders by text message and/or email upon request. Participants were contacted via email upon missing two consecutive daily surveys to maintain adherence to the study protocol and problem solve technical issues as needed. In addition, we included a short thank-you video at the end of the surveys midway through the study to encourage continued participation. Compensation was provided in the form of raffle entries toward one of ten US$50 Amazon gift cards. Participants were given 5 entries for completing the baseline survey, 1 entry per daily survey completion, and an additional 50 entries upon completing ≥ 85% of the daily surveys.

A total of 28,027 daily surveys were completed by participants (M = 54.42 days, Mdn = 64 days, SD = 21.46 days). Over half of participants (51%) completed more than 85% of the daily surveys, and most participants (87%) completed more than 30% of the daily surveys. Survey completion rates were negatively associated with trait (i.e., mean value across the study period) depressed mood (r = −.25) and loneliness (r = −.21), and cisgender women (M = 56.3, Mdn = 64, SD = 18.5 days) completed more surveys than cisgender men (M = 50.4, Mdn = 62.5, SD = 23.8 days).

Measures

Loneliness Loneliness was measured using two items designed for the purpose of this study and adapted from the UCLA Loneliness Scale (Russell, 1996). Each item was designed to prioritize face validity and, given space limitations inherent in daily survey research, to capture central aspects of loneliness. Participants responded to the two items “I felt lonely today” and “I felt left out today” using a Likert scale ranging from 0 (Not at all) to 10 (Extremely). A total score was computed by taking the mean across both items such that higher scores reflected greater loneliness. Coefficient ω indicated high between-person reliability (ω = .93) and moderate within-person reliability (ω = .65).

Social Interaction Quantity, Perceived Responsiveness, and Vulnerable Self-disclosure Social interaction quantity and quality were measured using four items designed for the purpose of this study to capture the interpersonal processes described by the Intimacy Process Model (Reis & Shaver, 1988). Participants responded to the following items, each with their own unique set of response options: “How much social interaction did you have today?” (social interaction quantity; 0 = None at all, 10 = A lot); “I felt understood/cared for by others today” (perceived responsiveness; 0 = Not at all, 10 = Extremely); “I expressed my true feelings to others today” (vulnerable self-disclosure; 0 = Not at all, 5 = A moderate amount, 10 = Very much so).

Data Analytic Strategy

Linear mixed models (i.e., multi-level models, hierarchical linear models) were used to address our main study aims. Mixed models allow for the disaggregation of within-person (i.e., state) and between-person (i.e., trait) effects (Enders & Tofighi, 2007; Wang & Maxwell, 2015) and thus allowed us to examine the effect of social interaction on loneliness and depressed mood independent of person-level averages. Mixed models also enable the estimation of person-level heterogeneity around fixed effects, which is particularly important in light of the fact that causal processes in psychology, including social processes, are not homogenous (Bolger et al., 2019; Hamaker, 2012).

The effect of state (i.e., time-varying) social interaction quantity, vulnerable self-disclosure, and perceived responsiveness was modeled by subtracting each individual’s observations from their average value across the EMA period (i.e., person-mean centering). The effect of trait (i.e., time-invariant) social interaction quantity, vulnerable self-disclosure, and perceived responsiveness was modeled by subtracting each individual’s average value across the EMA period from the sample average (i.e., grand-mean centering). Given their known associations with loneliness and depressed mood (Hawkley et al., 2020; Kessler & Bromet, 2013; Nolen-Hoeksema, 2001), we also included indicators of age, gender (0 = cisgender man), income, living arrangement (0 = lives alone), and relationship status (0 = single) in each model. Moreover, because these data were collected during the COVID-19 pandemic, and pandemic-related fears are associated with mental well-being (Brooks et al., 2020), we also included state and trait COVID-19 anxiety as a covariate in each model. Finally, the lagged outcome (i.e., previous day depressed mood/loneliness) was included in each model to control for stability in mood across days.

We used an iterative model building strategy wherein separate models were first constructed for each effect of social interaction quantity, vulnerable self-disclosure, and perceived responsiveness prior to estimating a full model that included all of these constructs. This strategy allowed us to identify changes in parameter estimates at each step, examine changes in model fit, and diagnose issues that may have occurred (e.g., with convergence). First, an “empty” intercepts-only model was estimated and subsequently compared against a model that included fixed effects for time (i.e., study day; 1–75) and weekend (0 = weekday, 1 = weekend). Then, time-varying and person-level variables were added sequentially, allowing us to carefully examine model performance as increasing complexity was introduced. Next, a random effect of each social interaction variable was added, which represents the unique effect for each individual around the average effect. The effect of the previous day’s depressed mood/loneliness was then added to control for stability in affect over time. A final model was estimated using restricted maximum likelihood estimation by including all three social interaction variables in the final model, testing all possible pairwise interactions among these variables, and subsequently estimating a model wherein only statistically significant interactions were included. A first-order autoregressive structure was placed on the residuals to account for the repeated measurements over time. All models were estimated using the “nlme” package (version 3.1-152; Pinheiro et al., 2020) for R (version 4.1.1; R Core Team, 2021). The R code for all analyses is included in the online Supplementary Material.

Bivariate Associations

Bivariate correlations are presented in Table 2. Small-to-moderate state (i.e., person-mean centered) associations were found between depressed mood/loneliness and social interaction quantity, vulnerable self-disclosure, and perceived responsiveness. Small-to-moderate associations were also observed between trait (i.e., grand-mean centered) depressed mood/loneliness and social interaction quantity, vulnerable self-disclosure, and perceived responsiveness. These results suggest that, in general, individuals who have more social interaction, engage in more vulnerable self-disclosure, and perceive greater responsiveness from others are less lonely and depressed and that, regardless of one’s trait level of these variables, daily increases in social interaction, vulnerable self-disclosure, and perceived responsiveness are associated with lower depressed mood and loneliness that same day.

OPEN IN VIEWER

Table 2.

Descriptive Statistics and Correlation Matrix of Primary Outcomes and Predictors.

	M (SD)	ICC	1	2	3	4	5
1. Depressed mood	2.69 (1.61)	.52		.74	−.23	−.33	−.35
2. Loneliness	2.39 (1.71)	.53	.53		−.20	−.28	−.37
3. Social interaction quantity	4.31 (1.37)	.33	−.22	−.24		.54	.58
4. Vulnerable self-disclosure	5.01 (1.88)	.49	−.12	−.13	.32		.78
5. Perceived responsiveness	5.34 (1.70)	.47	−.27	−.28	.33	.43

Moderate state and trait associations were also observed among all three social interaction variables. This validates the central dyadic process proposed by the IPM (i.e., the exchange of self-disclosure and responsiveness) and additionally suggests that greater social interaction quantity may bring about increased opportunity for engagement in vulnerable self-disclosure and perceived responsiveness. Nevertheless, these associations also raise the possibility that social interaction quantity and quality may not be independently associated with daily depressed mood and loneliness and necessitate the need for multivariate analyses.

Mixed Models

Model output for our primary analyses can be found in Table 3. For the full model output, see the online Supplementary Material.

OPEN IN VIEWER

Table 3.

Effects of State and Trait Social Interaction Quantity, Vulnerable Self-disclosure, and Perceived Responsiveness on Depressed Mood and Loneliness.

	Depressed mood			Loneliness
	B	SE	95% CI	b	SE	95% CI
Within-person effects
Social interaction quantity	−0.14	0.01	−0.15 to -0.12	−0.14	0.01	−0.16 to -0.13
Vulnerable self-disclosure	0.01	0.01	−0.01 to 0.02	−0.01	0.01	−0.02 to 0.01
Perceived responsiveness	−0.18	0.01	−0.19 to -0.16	−0.19	0.01	−0.21 to -0.17
Between-person effects
Social interaction quantity	−0.07	0.04	−0.14 to 0.01	−0.02	0.04	−0.10 to 0.05
Vulnerable self-disclosure	−0.06	0.03	−0.12 to 0.01	0.03	0.04	−0.04 to 0.10
Perceived responsiveness	−0.12	0.04	−0.20 to -0.05	−0.26	0.04	−0.35 to -0.18

Variance components
	1	2	3	4	1	2	3	4
1. Intercept	0.91				1.05
2. Social interaction quantity	−0.52	0.10			−0.58	0.13
3. Vulnerable self-disclosure	−0.11	−0.24	0.10		0.003	−0.20	0.11
4. Perceived responsiveness	−0.41	0.45	−0.05	0.12	−0.51	0.40	−0.15	0.15

Depressed Mood

A positive lagged effect of depressed mood was observed (b = 0.26, 95% CI = 0.25–0.27), indicating relative stability in depressed mood from one day to the next. Lower depressed mood was observed on weekends compared to weekdays (b = −0.11, 95% CI = −0.15 to −0.07). Transgender and gender non-binary participants reported greater depressed mood than cisgender participants (b = 0.35, 95% CI = 0.01–0.70), as did those with higher trait (b = 0.24, 95% CI = 0.21–0.27) and state (b = 0.11, 95% CI = 0.10–0.13) anxiety about COVID-19.

Social interaction quantity, vulnerable self-disclosure, and perceived responsiveness were differentially associated with depressed mood, and the effect of each differed at the state and trait levels of analysis. The effect of trait social interaction quantity was small and non-significant (b = −0.07, 95% CI = −0.14 to 0.01); however, state social interaction quantity was negatively associated with depressed mood (b = −0.14, 95% CI = −0.15 to −0.12), an effect that was slightly attenuated on days where individuals perceived higher responsiveness than usual (Figure 1a; b = 0.01, 95% CI = 0.003–0.01). There was no main effect of state (b = 0.01, 95% CI = −0.01 to 0.02) or trait (b = −0.06, 95% CI = −0.12 to 0.01) vulnerable self-disclosure on depressed mood; however, a significant interaction emerged such that state vulnerable self-disclosure was positively associated with depressed mood for individuals higher in trait social interaction quantity (Figure 1b; b = 0.02, 95% CI = 0.002–0.03), lower in trait vulnerable self-disclosure (Figure 1c; b = −0.02, 95% CI = −0.03 to −0.01), and on days where individuals perceived greater responsiveness than usual (Figure 1d; b = 0.01, 95% CI = 0.003–0.01). State vulnerable self-disclosure was negatively associated with depressed mood for individuals high in trait vulnerable disclosure (Figure 1c). A relatively strong effect of state (b = −0.18, 95% CI = −0.19 to −0.16) and trait (b = −0.12, 95% CI = −0.20 to −0.05) perceived responsiveness was observed such that feeling understood and cared for at both the state and trait levels is associated with lower depressed mood. ¹

OPEN IN VIEWER

Figure 1.

Marginal effect plots of the significant interactions in the depressed mood model. Note. Scores on the x-axis are represented in standardized units. Lines are represented by (A) state perceived responsiveness; (B) trait social interaction quantity; (C) traitvulnerable self-disclosure; (D) state perceived responsiveness.

Random slope estimates for each of the state social interaction variables suggested that there was meaningful person-level heterogeneity around the fixed effects. To estimate the range of effects that pertained to most participants, we calculated the middle 68% of effects, corresponding with roughly one SD below and above the average (i.e., fixed) effect estimate. The effect of state social interaction quantity ranged from −0.24 to −0.04 for most participants, suggesting that engaging in more social interaction than usual was more strongly negatively associated with depressed mood for some individuals than others (who experienced little-to-no effect). Similarly, the effect of state perceived responsiveness ranged from −0.30 to −0.05 for most participants. As expected, given its near-zero fixed effect, the random slope estimates of state vulnerable self-disclosure ranged from −0.09 to 0.10, suggesting that engaging in more vulnerable self-disclosure than usual was negatively associated with depressed mood for some individuals and positively associated for others.

Loneliness

A positive lagged effect of loneliness was observed (b = 0.26, 95% CI = 0.25–0.27), indicating relative stability in loneliness from one day to the next. Loneliness was lower on weekends compared to weekdays (b = −0.11, 95% CI = −0.15 to −0.07), for older participants (b = −0.01, 95% CI = −0.01 to −0.002), and for partnered versus single participants (b = −0.31, 95% CI = −0.53 to −0.08). Loneliness was higher for those with greater trait (b = 0.20, 95% CI = 0.16–0.23) and state (b = 0.09, 95% CI = 0.08–0.10) anxiety about COVID-19.

The pattern of effects for social interaction quantity, vulnerable self-disclosure, and perceived responsiveness on loneliness was similar to the effects of these variables on depressed mood. The effect of trait social interaction quantity was small and non-significant (b = −0.02, 95% CI = −0.10 to 0.05), however state social interaction quantity was negatively associated with loneliness (b = −0.14, 95% CI = −0.16 to −0.13), an effect that was stronger on days where individuals engaged in more vulnerable self-disclosure (Figure 2a; b = −0.01, 95% CI = −0.01 to −0.004) and perceived less responsiveness (Figure 2b; b = 0.01, 95% CI = 0.001–0.01) than usual. The effect of state social interaction quantity was also slightly weaker for individuals higher in trait perceived responsiveness (Figure 2c; b = 0.01, 95% CI = 0.004–0.02). Neither trait (b = 0.03, 95% CI = −0.04–0.10) nor state (b = −0.01, 95% CI = −0.02 to 0.01) vulnerable self-disclosure was associated with loneliness, however a significant interaction emerged such that state vulnerable self-disclosure was negatively associated with loneliness for individuals higher in trait vulnerable self-disclosure (Figure 2d; b = −0.02, 95% CI = −0.03 to −0.01) and on days where individuals perceived less responsiveness than usual (Figure 2e; b = 0.01, 95% CI = 0.003–0.01). State vulnerable self-disclosure was negatively associated with loneliness for individuals high in trait vulnerable disclosure (Figure 2d). A relatively strong effect of state (b = −0.19, 95% CI = −0.21 to −0.17) and trait (b = −0.26, 95% CI = −0.35 to −0.18) perceived responsiveness was observed such that feeling understood and cared for at both the state and trait levels is associated with lower depressed mood.

OPEN IN VIEWER

Figure 2.

Marginal effect plots of the significant interactions in the loneliness model. Note. Scores on the x-axis are represented in standardized units. Lines are represented by (A) state vulnerable self-disclosure; (B) state perceived responsiveness; (C) trait perceived responsiveness; (D) trait vulnerable self-disclosure; (E) state perceived responsiveness.

Significant person-level heterogeneity was observed for the effects of state social interaction quantity, vulnerable self-disclosure, and perceived responsiveness on loneliness. The effect of state social interaction quantity, vulnerable self-disclosure, and perceived responsiveness ranged from −0.27 to −0.02, −0.11 to 0.10, and −0.34 to −0.04, respectively, for most (i.e., 68%) participants.

Exploratory Analyses

The effects of social interaction quantity and quality were similar, although not identical, for depressed mood and loneliness, raising the possibility that the observed effects may be a function of the covariance between these two outcomes. Indeed, a robust association between loneliness and depression has been found in previous research (e.g., Cacioppo et al., 2010; Hawkley et al., 2020; Lim et al., 2016). To explore this possibility, we estimated mixed models with state and trait loneliness and depressed mood included as predictors.

Loneliness was strongly associated with depressed mood at both the state (b = 0.43, 95% CI = 0.42–0.44) and trait (b = 0.46, 95% CI = 0.41–0.51) levels. When loneliness was included in the model, the effect of state social interaction quantity was attenuated (b = −0.08, 95% CI = −0.09 to −0.07) but remained statistically significant. The effect of state perceived responsiveness in this model was also attenuated (b = −0.09, 95% CI = −0.11 to −0.08), and the effect of trait perceived responsiveness became non-significant (b = 0.01, 95% CI = −0.07 to 0.08). Interestingly, a small but significant effect of trait vulnerable self-disclosure was observed in this model (b = −0.09, 95% CI = −0.15 to −0.03).

A different pattern of results emerged for the effect of social interaction quantity and quality on loneliness when depressed mood was included in the model. Depressed mood was strongly associated with loneliness at both the state (b = 0.44, 95% CI =0.42–0.45) and trait (b = 0.53, 95% CI = 0.47–0.59) levels of analysis. Similar to the model for depressed mood, the effect of state social interaction quantity was attenuated (b = −0.08, 95% CI = −0.10 to −0.07) but remained statistically significant. However, the effects of state (b = −0.12, 95% CI = −0.13 to −0.10) and trait (b = −0.20, 95% CI = −0.28 to −0.12) perceived responsiveness both remained significant in this model. Moreover, trait vulnerable self-disclosure was positively associated with loneliness (b = 0.08, 95% CI = 0.01–0.14) such that individuals who engaged in more vulnerable self-disclosure on average were lonelier.

Limitations and Future Directions

There are several strengths of the current study, including the intensive longitudinal daily diary design, the differentiation of within-person and between-person effects, and the identification of heterogeneity in the observed effects via random slopes analysis. These strengths, however, come with notable limitations, and caution should be exercised in interpreting the results.

It is possible that our findings are a function of our sampling rate (i.e., daily), and that different results would be obtained if we sampled at shorter or longer time intervals. Thus, several important questions remain unanswered, such as whether the benefits of social interactions are experienced immediately (i.e., during the interaction) or if there is a delayed or even cumulative effect throughout the course of the day. For example, research suggests that laughter during social interactions predicts increased closeness, enjoyment, and positive emotionality in subsequent interactions that same day (Kashdan et al., 2014) and that enacted responsiveness predicts immediate (i.e., post-interaction) perceived responsiveness (Maisel et al., 2008). It is thus possible that the results of the current study generalize to the level of each individual interaction; however, future research is needed to test this hypothesis.

Importantly, we are limited in our ability to draw causal conclusions given the non-experimental design of this study. Although we offered what we consider to be the most plausible interpretations given our modeling strategy (i.e., each outcome as a residualized change score; Castro-Schilo & Grimm, 2018), theory (IPM; Reis, 2007), and the existing literature, we cannot rule out the possibility of reverse causality (i.e., depression and loneliness bringing about changes in our social interactions) or a “third variable” that accounts for changes in depressed mood, loneliness, and our social interactions. For example, depressed individuals tend to be more socially isolated than non-depressed individuals (Brown et al., 2011; Nezlek et al., 2002) and may be biased toward interpreting threat rather than acceptance in social interactions (Roberts et al., 2010). Lonely individuals may similarly exhibit a bias toward social threat (Bangee et al., 2014; Cacioppo, Balogh, & Cacioppo, 2015; Cacioppo, Bangee, et al., 2015). Future research would benefit from using an experimental design to establish causality of these effects.

Data for the current study were collected at the beginning of the COVID-19 pandemic in the United States. It is thus possible that our findings are unique to the structure of social interactions during the pandemic and may not generalize to normal (i.e., non-pandemic) periods of social life. Although virtual forms of communication (i.e., phone, video-chat, instant messaging) are sufficient to facilitate interpersonal connection, in-person interactions may nevertheless be superior (Sherman et al., 2013). We did not differentiate between in-person and virtual forms of communication in the current study, and it is likely that many interactions occurred virtually due to the pandemic. Furthermore, participants may have interpreted the social interaction quantity item differently (e.g., how many people they interacted with, how much time they spent interacting with others). Future research should improve precision of this item, focus on replicating the current findings in a non-pandemic context, and explore whether the effects of social interaction quantity and quality on mental health are moderated by the use of technology (or lack thereof).

Generalizability of the current findings may also be limited by the nature of our sample, the majority of whom were white, cisgender women, in a romantic relationship, and lived in urban/suburban settings. It is possible that our findings reflect cultural processes that pertain to our sample and do not generalize well to individuals with different demographic characteristics. For example, evidence suggests that culture may influence the degree to which individuals experience somatic versus psychological symptoms of depression (Parker et al., 2001). Given that the items assessing depressed mood in the current study pertained uniquely to psychological symptoms of depression, our findings may not generalize to the experience of those with more heavily somatic presentations. Future research should attempt to replicate these findings in more culturally diverse samples in order to better understand the potential influence of culture on the association between social interactions and mental health.

Finally, the degree to which the current results reflect clinically meaningful processes is unknown given that these data were derived from a community sample of adults, the majority of whom were not clinically depressed or otherwise seeking psychiatric treatment. Research suggests that subthreshold depressive symptoms are associated with clinically meaningful levels of distress and dysfunction (Cuijpers & Smit, 2004; Lewinsohn et al., 2000), and that because depression is most accurately conceptualized as a dimensional construct (Kendler & Gardner, 1998; Slade, 2007), predictors of change along that dimension are important etiological factors (Nelson et al., 2017; Wichers, 2014). Nevertheless, future research would benefit from exploring these processes in clinical samples.