A Day in the Life: Couples’ Everyday Communication and Subsequent Relationship Outcomes

Abstract

Understanding how communication processes contribute to well-functioning versus distressed couple relationships has relied largely on brief, laboratory-based conversations. Harnessing technological advancements, the current study extends the literature by capturing couples’ naturalistic communication over one full day at Time 1 (T1). This study tested associations between data-driven categories of couple communication behaviors and relationship outcomes (i.e., relationship aggression, satisfaction, and dissolution) at Time 2 (T2), approximately one year later. Emerging adults in different-gender dating couples (n = 106 couples; 212 individuals; M age = 22.57±2.44; M relationship length = 30.49 months±24.05; 72.2% non-White) were each provided a smartphone programmed to audio record approximately 50% of a typical day. Interactions between partners were transcribed and coded for location, activity, affect, and a range of positive and negative communication behaviors for each partner. Even after controlling for T1 assessments of the relevant outcome, one’s own hostility and one’s partner’s hostility at T1 were each positively associated with T2 relationship aggression and negatively associated with T2 relationship satisfaction. One’s own withdrawal at T1 was positively associated with T2 relationship aggression perpetration, whereas one’s partner’s withdrawal was negatively linked to relationship satisfaction at T2. One’s own playfulness, unexpectedly, was linked to lower subsequent relationship satisfaction. Withdrawal increased the likelihood of relationship dissolution, whereas warmth and playfulness decreased the likelihood of dissolution. The relevance of couples’ ordinary, everyday communication for meaningful relationship outcomes is discussed.

Direct observation has been a mainstay in generating new knowledge about couple interaction, but has been largely based on standardized lab-based discussions, leaving large gaps in types of interactions sampled. Technological innovations make it possible to capture everyday couple interactions in a variety of contexts (Bulling et al., 2023; Reblin et al., 2018). Such real-world phenomena putatively are the key to understanding relationship maintenance, relationship problems, and can aid in the identification of dyadic intervention targets. Naturalistic observation outside the lab provides a window into the rhythm and contours of a couple’s day, including an unlimited range of topics, from the important and serious to the small-talk, logistical, or lighthearted, as well as varying activities, such as completing chores or simply “hanging out”. The present study leverages mobile phone audio recordings for automated, passive capture of naturalistic interactions across “a day in the life” of young adult, different-gendered couples. Our study provides descriptive information about the nature of couples’ daily interactions and the feasibility of collecting such ambulatory data. We further investigate whether naturalistic communication behaviors between partners are meaningfully linked to concurrent and subsequent relationship aggression and satisfaction, as well as to relationship dissolution.

Studying Couple Communication

A large number of studies have used observational data to study couple communication patterns, understand group differences (e.g., between distressed and non-distressed couples), and identify couple dynamics associated with individual characteristics, such as physical and mental health (for overviews, see Gottman & Notarius, 2000; Friedlander et al., 2019; Heyman et al., 2014; Kerig & Baucom, 2004; Kiecolt-Glaser & Newton, 2001; Margolin et al., 1998). Sampling communications has contributed to both hypothesis-generation and hypothesis-testing about the course of relationships and the ways that they contribute to individual well-being. Behavioral observations of couples typically involve auditory and/or visual signals. In most laboratory-based observations, couples are instructed to discuss ongoing disagreements (e.g., see Heyman, 2001; Woodin et al., 2011), although other lab-based discussions involve the disclosure of personal stressors, worries, losses, and self-improvement goals (e.g., Cutrona et al., 1997; Overall et al., 2010; Khalifian & Barry, 2021; Margolin et al., 2022; Pasch & Bradbury, 1998). Regardless of topic, such discussions assume that short (e.g., 10–15 minute), prompted conversations are representative of spontaneously occurring interactions on these topics. Advantages of lab-based discussions include their efficiency and standardization, which facilitates comparisons across couples. The key drawback, however, is generalization since couples do not typically sit down for time-limited face-to-face discussions about disagreements or personal stressors. Moreover, laboratory-based discussions may also underestimate negativity in conflicts (Heyman, 2001) and limit opportunities for certain conflict behaviors, such as withdrawal.

As an alternative, naturalistic interactions samples outside the lab can capture snippets of conversations interwoven with couples’ daily activities. Naturalistic data, which are much more resource-intensive to collect and process, reflect the assumption that the ordinary moments of life vary in meaningful ways across couples and hold significance for how individuals experience themselves, their partner, and the relationship more generally. Thus, studying experiences as they naturally occur improves ecological validity by offering insight into the likelihood and impact of everyday behaviors (Repetti et al., 2013). However, questions remain about whether observational data in the real-world generate enough behaviors of interest for hypothesis testing and whether momentary, and at times subtle, behaviors cluster into meaningful units that shed light on theoretically relevant relationship constructs (Heyman et al., 2014).

The earliest attempts to capture life as it is lived relied on quasi-ethnographic methods, in which human observers, stationed inconspicuously in families’ homes, coded family interactions in real-time (Patterson, 1982). Alternatively, videographers followed family members around the home to record couples’ re-enacted, recent conflicts (Burman et al., 1993) or family routines (McNeil & Repetti, 2021). These methods reveal how interactions move across time and space, but the impact of a third-party’s presence is difficult to gauge. Other methods included placing audio-recording equipment in families’ homes, preset to activate at intervals unbeknownst to the family (Christensen & Hazzard, 1983), or having couples spend a weekend in a studio apartment laboratory equipped with cameras (Gottman & Driver, 2005). As an alternative to human-operated recordings or complicated equipment set-ups, wearable microphones have been used to capture real-life interaction. Specifically, an important breakthrough in capturing naturalistic interaction was the development of the Electronically Activated Recorder (EAR; Mehl & Robbins, 2012), a portable audio device that intermittently records brief, 6–50 second clips of ambient sound in daily life. When applied to couple dynamics, specifically couples living with a cancer diagnosis, EAR data showed that negative emotion, anger words, and pronoun use are associated with dyadic adjustment (Karan et al., 2017; Robbins et al., 2014).

Communication and Outcomes: Relationship Aggression, Satisfaction, and Dissolution

Despite the aforementioned generalizability concerns surrounding lab-based discussions, behaviors observed during these interactions have relevance for global relationship outcomes, such as aggression, satisfaction, and relationship dissolution. Lab-based behaviors during conflict discussions, including high hostility, low reciprocity and warmth, and poor problem-solving, have been linked to aggression in romantic relationships (e.g., Capaldi et al., 2003; Margolin et al., 1988; Sommer et al., 2019). Even in discussions involving non-conflictual topics, such as planning a date, aggressive partners exhibit more negative and fewer positive behaviors (Daspe et al., 2022). Outside of the lab, time-limited home discussions (Hammett et al., 2021), as well as home reenactments of prior conflicts (Burman et al., 1993), similarly evoke more anger, contempt, and aversive behaviors in aggressive couples. To our knowledge, however, other than a prior study on pronoun use in daily life (Timmons et al., 2021), naturalistic observations of couples have not yet addressed whether couples who have experienced relationship aggression communicate in distinctive ways in everyday conversations. Moreover, naturalistic conversations have not yet been examined for predicting later aggression.

Identifying communication factors that contribute to relationship quality has been an important goal of couple research, with considerable support for communication-based theories that highlight behavioral skills deficits. In lab-based conflict discussions, distressed couples show more negative and fewer positive behaviors, with similar findings in support-provision discussions (for reviews, Gottman & Notarius, 2000; Heyman, 2001). Woodin’s (2011) meta-analysis further suggests that high-intensity positive and negative conflict behaviors are robustly linked to relationship satisfaction. Longitudinal links between observed communication and later relationship satisfaction are more mixed, however. Although multiple meta-analyses reported small to moderate associations between communication and later relationship quality (Kanter et al., 2022; Karney & Bradbury, 1995), other studies reveal inconsistencies. For example, a four-wave study of newlyweds showed consistent cross-sectional links between communication and relationship satisfaction, but found limited longitudinal support (Lavner et al., 2016). In premarital couples, negative, but not positive, observed communications were associated with lower relationship adjustment and, unexpectedly, negative communications were associated with less steep declines in satisfaction over time (Markman et al., 2010).

The important question of whether certain communication behaviors forecast relationship dissolution has generated mixed conclusions. A recent meta-analytic study concluded that both self-reported and observed hostility are common predictors of relationship dissolution (Kanter et al., 2022). In an earlier review, communication behaviors accounted for relatively small effects on relationship dissolution (Karney & Bradbury, 1995). Gottman and colleagues (e.g., Gottman et al., 1998; Gottman & Levenson, 2002) reported robust findings that observed emotions can predict divorce, whereas others have not replicated these findings (Kim et al., 2007; Markman et al., 2010). Studies have yet to examine whether mundane, everyday interactions that occur outside of staged lab-based conflicts may, over time, contribute to relationship dissolution.

Current Study

The present study is based on the premise that to better understand relationships, we need to know more about what couples do outside the lab in unstructured conversations in the midst of their natural daily activities (Heyman, 2001). We use technological advancements to “listen in” on spoken communications and ambient sounds in different-gender young couples’ everyday lives. With smartphones programmed to audio record at set intervals, we captured approximately 50% of couples’ interactions across one typical day. Audio recordings were then transcribed and coded for location, activity, and affective and communication behaviors between partners.

The first aim of this study is to depict the range and nature of young couple interactions and activities. We provide descriptive information on coded data, in addition to information about participants’ compliance, reactivity to data collection procedures, and representativeness of the day. For aim 2, given the lack of prior information about naturalistic observation with couples, we identify categories of communication behavior beyond general categories of positivity and negativity. Following strategies used in other coding systems to reduce the number of codes examined (e.g., Heyman et al., 2021), we use a data-driven approach to extract broader constructs from the more microanalytic individual codes. We also examine how the identified categories of behavior are cross-sectionally linked to each other and to relationship aggression and satisfaction. Based on prior lab research, we anticipate that relationship aggression will be associated with negatively-valenced constructs whereas satisfaction will be associated with both low-negative and high-positive constructs. Aim 3 builds on assumptions that everyday communications have a cumulative effect on relationship quality and stability. We investigate whether naturalistic daily interactions are related to relationship functioning one year later, specifically testing whether each partner’s use of the communication constructs that emerge in Aim 2 are linked to subsequent relationship aggression, satisfaction, and dissolution. In light of prior longitudinal research that negative, compared to positive, communications are more consistently associated with later relationship outcomes, we hypothesize that negative constructs will be linked to lower relationship satisfaction, higher aggression, and relationship dissolution. We do not have specific predictions for positively-valenced behaviors as in-lab findings may not generalize given the potential for a broader range of observable positive behaviors in naturalistic interactions. We also do not have predictions about gender differences given mixed findings from lab-based studies and lack of prior information from naturalistic observations.

Method

Overview

Couples completed Time 1 (T1) self-report online surveys and participated in a laboratory visit as part of a larger study (Margolin et al., 2022). After the laboratory visit, couples were invited to participate in the home data project (see Timmons et al., 2017 for a full description), which involved a laboratory meeting to receive the recording devices, data collection over an entire day outside the lab, and a second visit for exit interviews and to return devices. Devices captured naturalistic audio recordings, physiological measurements, and hourly self-reports (Timmons et al., 2019; 2021). Data collection was scheduled on a day when partners agreed to spend at least five hours together. Per participants’ self-reports, couples spent an average of 11.5 hours together (SD = 2.68, Mode = 13). At Time 2 (T2), approximately one year later, experimenters invited participants to complete a brief online follow-up survey and to participate in laboratory procedures. The data used in this paper include T1 self-report survey data, T1 audio-recordings from the day-long procedures as well as questionnaires from the return visit the following day, and T2 self-reported relationship outcomes. This project received approval from the University of Southern California Institutional Review Board.

Participants

Of the sample for the larger project (n = 121 couples), 106 different-gender, young adult couples (n = 212 individuals; Mage = 22.57; SD = 2.44) participated in the day-long study. Given our small number of same-gender couples (n = 3), we focused this initial paper on observable behaviors in different-gender couples. To be eligible for the larger study, couples needed to be together for a minimum of two months and at least one partner needed to be between the ages of 18–25. Couples were largely recruited via online and community advertisements, with 24.5% of participants recruited from a long-term longitudinal study on associations between family aggression and adolescent development (Margolin et al., 2010). Of the 106 participating couples, median relationship length was 22.5 months (M = 30.49 months, SD = 24.05) and 43.4% of couples lived together. The majority of participants (74.1%) were employed and 55.2% were students. Self-reported ethnicity was: 27.8% White, 23.6% Hispanic/Latinx, 16.0% multiracial, 15.6% Black/African American, 13.2% Asian/Asian American, and 3.8% other. At T2, the sample consisted of 186 individuals across 101 couples. Attrited participants did not differ from retained participants in age, relationship length, or cohabitation status.

Home Data Collection Procedures

Experimenters met with couples at 10:00AM on the day of data collection to explain procedures, obtain informed consent, and loan each partner a Nexus 5 smartphone that we programmed to unobtrusively and automatically record audio (via the application RecForge II) for 3-minute intervals every 12 minutes during each partner’s waking hours (see Timmons et al., 2017). With recordings alternating between partners’ devices, procedures were designed to capture 6 minutes of every 12-minutes across both smartphones, such that approximately 50% of each waking hour was audio-recorded between 10:00 AM until 3:00 AM (or when participants powered off devices at bedtime). Participants were unaware when the smartphones were recording but were instructed to mute the device for their privacy and in the presence of unconsented individuals. Per Robbins (2017), to further protect the privacy of bystanders, researchers instructed participants to wear a button that stated, “This conversation may be recorded” and provided participants with informational handouts on the study to share with individuals with whom they had extended contact. The following day, participants returned the smartphones, received $100 compensation, and completed an exit interview regarding their activities and their experiences with the data collection procedures.

Measures

Behavioral Observations

Behavioral observations were made with a multifaceted coding system that included codes for location, activities, presence of others, and communication tone and content between partners. Methods were informed, in part, by the EAR coding (Mehl & Robbins, 2012; Mehl, personal communication, 2014), along with our lab’s prior couple interaction coding systems (e.g., Burman et al., 1993). Coding was completed based on contextual and verbal information. The coding system included dichotomous (0 = absent/no, 1 = present/yes) codes for location (11 codes; e.g., home, vehicle), activity (34 codes; e.g., singing, completing chores), and presence of others who might be the recipient of spoken communication (9 codes; e.g., family member, friend). Couple communication codes described the content of the speech between partners (18 codes; e.g., insulting, validating), and affect codes described tone of voice (9 codes; e.g., irritated, playful). When partners were apart and not speaking to each other, raters only coded location, activity, and presence of others. Communication behaviors and tone of voice were coded only for couple interactions, either when partners were physically together or speaking by phone. For both ethical/legal reasons and due to the substantive questions addressed here, recorded speech directed towards or from other recipients (e.g., friends, family, strangers), was not coded. For each 3-minute recording, each code was rated on a 3-point scale (0 = absence of behavior/tone, 1 = presence of behavior/tone with low frequency or intensity, and 2 = presence of behavior/tone with high frequency or intensity) for each partner. Average scores were calculated by taking the mean rating across all available audio files over the full day.

Dialogue from all audio files was manually transcribed, then verified for accuracy by two research assistants. Raters received at least 15 hours of initial training, including watching asynchronous instructional videos, attending weekly meetings, and coding several audio files of couple interaction for practice. Upon completing the training, raters needed to achieve at least 80% reliability on a “coding test” of three complex audio recordings before progressing to data coding. Raters coded individual three-minute audio files chronologically (i.e., alternating between recordings from each partner’s phone) and listened to each audio recording at least two times. Due to the difficulty of extracting the specific speech content within the ambient noise of everyday life, raters were instructed to read transcripts of the audio files as they coded the audio-recordings. To assess reliability, codable files from a 3-hour segment of each couple’s day, selected from the merged audio data across both partners’ phones, was coded by a second rater. Inter-rater reliability was calculated using Gwet’s agreement coefficient 2 (AC2; Gwet, 2008), to account for skewed distributions (i.e., low frequencies of behaviors). Reliability for all codes used in the substantive analyses were high (all ≥.90).

Representativeness and Reactivity of Data Collection

As part of the exit interview, participants independently completed a questionnaire designed for this study that assessed: (a) how typical their day was in terms of interactions with their partner, and (b) whether they changed their behaviors knowing that the recording may be activated. Responses were made on a scale from 0 (Not at all) to 4 (Extremely).

Relationship Aggression

Relationship aggression was measured via the 65-item How Partners and Friends Treat Each Other questionnaire (Bennett et al., 2011), which assesses physical, psychological, sexual, and electronic dating aggression. Participants completed the scale twice–reporting their own perpetration of aggression towards their partner, and then reporting their partner’s aggression towards them; response options ranged from Never happened (0), to Happened more than ten times (4). At T1, participants reported aggression that occurred within the past year, and at T2, participants reported aggression that occurred since their initial laboratory visit. Because aggression tends to be under-reported, we compared both partners’ reports (e.g., women’s reports of partner’s aggression and men’s reports of their own aggression) for each item and used the maximum report per item to calculate mean aggression perpetration across all items for each person (T1 and T2 α = .96 for women, .94 for men). Women reported greater aggression perpetration compared to men at T1 (women: M = .38, SD = .04; men: M = .30, SD = .03, p = .001), but not at T2 (women: M = .29, SD = .05; men: M = .28, SD = .03, p > .05).

Relationship Satisfaction

Relationship satisfaction at T1 and T2 was assessed via a mean score of the Quality of Marriage Index (QMI; Norton, 1983). Five items are assessed on a 7-point scale (1 = strongly disagree, 7 = strongly agree). The final item measures the degree of happiness in the relationship on a 10-point scale (1 = unhappy, 10 = perfectly happy). Cronbach’s α was ≥ .94 at T1 and T2 for both men and women. Women reported lower relationship satisfaction at T1 (women: M = 6.43, SD = .11; men: M = 6.67, SD = .08, p = .037), but not at T2 (women: M = 6.21, SD = .19; men: M = 6.16, SD = .17, p > .05).

Relationship Dissolution

At T2 follow-up, participants were asked whether their relationship had ended. In total, 22 couples (21.8%) had broken up and 78 remained together. One couple reported discrepant responses about relationship status and was thus dropped from dissolution analyses.

Analytic Plan

Aim 1 presents descriptive statistics on our coded audio data, including the percent of files in which partners were together and communicating. We created box and whisker plots to illustrate the range and median percentage of audio files across couples that were sampled across different locations and activities. For Aim 2, we conducted a principal components analysis (PCA) with oblimin rotation to reduce the 27 individual codes into more comprehensive communication components. Contemporaneous associations between the PCA-derived communication components and T1 relationship aggression and satisfaction are examined through Pearson correlations. For Aim 3, we examined how one’s own, and one’s partner’s communication components relate, a year later, to relationship aggression, relationship satisfaction, and relationship dissolution. Aggression and satisfaction models were conducted using linear mixed modeling with individuals nested within couple dyads, with the lme4 package and dissolution was examined via logistic regression in R. Aggression models included couples who had T2 aggression data, whereas satisfaction models were restricted to couples who did not break up. Relationship length and gender were included as covariates in all models. In addition, models controlled for T1 assessments of the relevant outcome (i.e., T1 relationship aggression for T2 aggression; T1 relationship satisfaction for T2 satisfaction and dissolution). We reported how we determined our sample size, all data exclusions, manipulations, and measures in the study. This study was not pre-registered. Anonymized datasets can be obtained upon request.

Results

Aim 1: Descriptive Information

Data collection yielded 11,856 three-minute audio files, which were pared down to 6,912 files for coding (See Supplemental Figure 1). On par with similar data collection methods (e.g., Macbeth et al., 2022), 9,537 files (80.4%) were usable after exclusions based on: (a) equipment failures [i.e., recording malfunctions (7.8%), duplicate files (0.7%), or static throughout the recording (2.5%)]; (b) recordings while still receiving instructions in the lab (3.8%); (c) uninformative noise (2.2%); or (d) muted audio (2.6%). To focus on dyadic constructs of interest, we further restricted all analyses to files in which partners were physically together and interacting (i.e., at least one word spoken), or apart and interacting (e.g., on the phone). All analyses here are based on the 6,912 files that met these inclusion criteria (72.5% of usable files). The average number of audio files per couple that met inclusion criteria was 65.21 (SD = 29.14, median = 67, range = 8 – 124), or approximately 195 recorded minutes of interaction. There were no gender differences in the average of recorded words spoken to one’s partner over the day (women: M = 4,302.4, SD = 2,765.5; men M = 4,205.9, SD = 2,752.8). Total word count per couple is unrelated to relationship length or whether the couple cohabitates (both r’s = .01).

The majority of participants (71.5%) indicated that the day of data collection was extremely or very typical of how they usually interact with their partner. Most (90.4%) also reported that they either did not or only slightly changed behaviors as a result of audio recording. Figure 1 provides box and whisker plots that depict median percentages of audio files that couples spent in various activities and locations over the day. Across couples, approximately half (50.2%) of audio files captured recordings in one or both partners’ homes, another 27.0% in cars or public transportation, 11.7% outdoors, 5.2% at restaurants/cafes, with other locations (stores, others’ homes) each accounting for less than 5%. Notably, only .6% of files were coded as “unknown” location. Activity data show that couples commonly spent time listening to music (29.2%), driving (23.3%), watching TV (18.2%), walking (10.2%), singing (8.6%), using devices (8.0%), and eating (8.0%), with a range of other activities each occurring < 5% of the time.

Figure 1.

Box and Whisker Plot Depicting Median Percentages of Audio Files that Couples Spent Doing Activities over the Day

Box and Whisker Plot Depicting Median Percentages of Audio Files that Couples Spent in Various Locations over the Day

Aim 2: Principal Components Analysis of Observations of Everyday Communication

We first examined the suitability of our data for PCA as a dimension reduction technique. Initial inspection of the correlations between communication codes (See Supplemental Table 1) revealed four codes (ignoring partner, depressed tone, collaboration, distracted by electronics) that did not correlate >.30 with any codes and thus were dropped. The overall Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was .78. Inspection of KMO for each individual code revealed one code (neutral tone) with an unacceptable KMO (<.5). After dropping this code, our sampling adequacy was .80 (‘meritorious’; Kaiser, 1974). Bartlett’s test of sphericity was statistically significant (p < .001), suggesting suitability of our data to PCA. Both a parallel analysis and visual inspection of the scree plot indicated a four component solution.

We employed an oblique oblimin rotation to aid interpretability and allow components to correlate. Two codes (defending, conflict) were eliminated that cross-loaded >.4 on two components. The resulting rotated factor solution revealed four components that each comprised at least three individual communication codes above a cut-off of .4 (See Table 1). Components had eigenvalues exceeding 1.0 and together explained 55% of the total variance. One component, hostility, reflected more overtly negative behaviors whereas vulnerability contained potentially positive (vulnerable) and negative or ambiguous (anxious tone, complaining) behaviors. The two positively-valenced constructs that emerged were warmth and playful communication. The withdrawal code was the only code that did not load highly on any component. Because of its theoretical importance (Gottman & Notarius, 2000), we include withdrawal as a stand-alone construct in our substantive analyses. See Supplemental Table 2 for descriptions of the codes comprising each component. Component scores were used for substantive analysis.

Table 1.

Principal Component Analysis of Coded Naturalistic Communication Behaviors

		Component Factor Loadings
Variable	1 Hostile	2 Vulnerable	3 Playful	4 Warm
Dismissing/Invalidating	.83	−.02	.01	.18
Hostile/Irritated Tone	.80	.17	−.05	−.11
Badgering and Baiting	.78	−.20	−.08	−.04
Insulting/Criticizing/Blaming	.76	.22	−.01	−.01
Whiny Tone	.52	−.15	.37	−.13
Interrupting	.48	.23	−.08	.19
Dominating Conversation	.42	.07	.13	.11
Expressing Vulnerability	.03	.80	−.01	.01
Serious Tone	.03	.77	−.12	.16
Anxious Tone	.05	.74	.12	−.06
Complaining	.24	.60	.14	−.09
Silly/Playful Tone	.08	−.08	.88	.04
Using Humor	−.11	.09	.85	−.10
Enjoying Interaction	−.07	−.07	.73	.26
Enthusiastic Tone	−.03	.23	.51	−.22
Being Engaged	.05	.24	.48	.24
Using Words of Affirmation	.05	.08	.00	.75
Warm/Supportive Tone	.07	−.18	.13	.70
Validating	−.08	.30	.01	.62
Withdrawing	.35	.06	−.05	−.31
Variance Explained	.18	.14	.14	.10

Note. These results represent the four-factor solution to the PCA with oblique oblimin rotation.

Table 2 presents correlations among study variables. The five communication constructs were moderately-to-highly correlated between partners (.69 for hostile, .43 for withdrawal, .83 for anxious/vulnerable, .68 for warm, and .86 for playful). Both own and partner’s hostility were positively associated with both partners’ withdrawal and vulnerability. Own withdrawal was positively linked to the partner’s vulnerability and negatively linked to their own and the partner’s warmth. Warmth was positively linked to both partners’ playfulness and vulnerability. Associations between communication and concurrent relationship characteristics show that both one’s own and the partner’s hostile, withdrawn, and vulnerable behaviors are positively associated with T1 aggression. Both one’s own and the partner’s vulnerability are negatively associated with T1 satisfaction. Table 2 also shows bivariate associations between T2 outcomes and communication behaviors, which are tested more rigorously in Aim 3.

Table 2.

Bivariate Correlations Among Study Variables

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
1. Hostility (Self)
2. Withdrawal (Self)	.36**
3. Vulnerability (Self)	.33**	.12
4. Warmth (Self)	.05	−.29**	.14*
5. Playful (Self)	.00	−.09**	.05	.17*
6. Hostility (Partner)	.69 **
7. Withdrawal (Partner)	.32**	.43 *
8. Vulnerability (Partner)	.33**	.15*	.83 **
9. Warmth (Partner)	.12	−.15*	.23**	.68 **
10. Playful (Partner)	−.01	−.08	−.01	.27**	.86 **
11. Aggression Perpetration (T1)	.32**	.17*	.24**	−.13	−.12	.28**	.19**	.23**	−.11	−.13
12. Aggression Perpetration (T2)	.28**	.35**	.07	−.06	−.03	.27**	.25**	.07	−.03	−.03	.43**
13. Relationship Satisfaction (T1)	−.14	−.14	−.25**	.10	.06	−.12	−.09	−.28**	.01	.11	−.27**	−.10
14. Relationship Satisfaction (T2)	−.26**	−.12	−.18*	.00	−.21**	−.21*	−.19*	−.11	−.07	−.14	−.13	−.06	.09
15. Relationship Dissolution	.12	.31**	.08	−.20**	−.18**						.25**	.32**	−.18*	--
16. Relationship Length	.08	−.03	.21**	−.10	−.02						.11	−.11	−.01	−.14	−.08

Note. *p < .05. **p < .01. Variables 1–10 are T1 communication constructs based on individual, standardized communication scores from the PCA (M = 0, SD = 1). Correlations between Self and Partner’s communication behaviors are depicted in bold. Intercorrelations for partners’ communication codes are not depicted since they are the same as Self. Relationship dissolution and relationship length are the same for Self and Partner and thus are presented only for Self. Relationship satisfaction at T2 was not measured for couples who ended their relationship. Correlations for relationship dissolution are Point-Biserial. All others are Pearson.

Aim 3: Everyday Communication Constructs and Relationship Outcomes at T2

Table 3 presents results for the models testing one’s own and the partner’s communication constructs on T2 relationship aggression and satisfaction. Due to the high correlations between partners on the coded communication behaviors, we ran separate analyses for self and partner’s communication. In line with our hypotheses, significant effects emerged for one’s own hostility and withdrawal, as well as the partner’s hostility on T2 aggression perpetration, even controlling for T1 aggression. Neither partner’s vulnerable, warm, and playful behaviors were significantly associated with T2 aggression. Results for T2 relationship satisfaction partially support our hypothesis; controlling for T1 satisfaction, both one’s own and the partner’s hostility, and only the partner’s withdrawal, predicted lower later satisfaction. One’s own playfulness was also associated with T2 satisfaction but, unexpectedly, in an inverse direction. Neither partners’ vulnerability nor warmth contributed to T2 satisfaction. Table 4 presents the logistic regressions for communication constructs and relationship dissolution, controlling for T1 relationship satisfaction. With relationship dissolution a joint outcome across both partners, we present the results for own communication and relationship dissolution; parallel analyses for the partner’s communication show the same pattern and nearly identical effect sizes. Withdrawal was positively associated with relationship dissolution, whereas both warm and playful behaviors were negatively associated with relationship dissolution. Neither hostile nor vulnerable behaviors significantly predicted dissolution.

Table 3.

Individual’s and Partner ’s Communication Behaviors and Subsequent Dating Aggression Perpetration and Relationship Satisfaction

		T2 Aggression Perpetration				T2 Relationship Satisfaction
	Effects Added to Model	Communication Behavior (Self)		Communication Behavior (Partner)		Communication Behavior (Self)		Communication Behavior (Partner)
		B	SE	B	SE	B	SE	B	SE
Hostile	Intercept	−.02	.18	.05	.18	.03	.26	−.09	.26
	T1 Dating Aggression Perpetration	40***	.08	.40***	.08	−	−	−	−
	T1 Relationship Satisfaction	-	-	-	-	.07	.09	.05	.09
	Relationship Length	−.16*	.07	−.16*	.07	−.11	.08	−.11	.08
	Gender	.04	.11	−.01	.11	−.05	.16	.04	.16
	Hostile Communication	.18 *	.07	.17 *	.07	−.28**	.09	−.21*	.09
Withdrawn	Intercept	.01	.17	.01	.18	−.02	.26	−.10	.26
	T1 Dating Aggression Perpetration	39***	.08	.43***	.08	-	-	-	-
	T1 Relationship Satisfaction	-	-	-	-	.09	.09	.09	.09
	Relationship Length	−.13*	.07	−.14*	.07	−.12	.08	−.12	.08
	Gender	.01	.11	.01	.11	−.02	.17	.02	.16
	Withdrawn Communication	.29***	.07	.11	.07	−.17	.13	−.29*	.13
Vulnerable	Intercept	.01	.18	−.00	.18	.02	.26	.02	.27
	T1 Dating Aggression Perpetration	45***	.08	.45***	.08	-	-	-	-
	T1 Relationship Satisfaction	-	-	-	-	.04	.09	.07	.09
	Relationship Length	−.14*	.07	−.15*	.07	−.10	.08	−.11	.08
	Gender	.02	.11	.02	.11	−.02	.17	.00	.17
	Vulnerable Communication	−.02	.07	−.01	.07	−.12	.08	−.05	.08
Warm	Intercept	−.01	.18	.00	.18	−.01	.27	.04	.27
	T1 Dating Aggression Perpetration	.44***	.08	.45***	.08	-	-	-	-
	T1 Relationship Satisfaction	-	-	-	-	.08	.09	.09	.09
	Relationship Length	−.15*	.07	−.15*	.07	−.13	.08	−.13	.08
	Gender	.03	.11	.02	.11	−.01	.17	−.03	.17
	Warm Communication	−.04	.07	−.00	.07	−.01	.08	−.08	.08
Playful	Intercept	−.01	.18	.01	.18	.07	.26	−.02	.26
	T1 Dating Aggression Perpetration	.45***	.08	.45***	.08	-	-	-	-
	T1 Relationship Satisfaction	-	-	-	-	.10	.09	.10	.09
	Relationship Length	−.15*	.07	−.15*	.07	−.12	.08	−.12	.08
	Gender	.02	.11	.01	.11	−.04	.16	.01	.17
	Playful Communication	.03	.07	.04	.07	−.21**	.08	−.14	.08

Note. *p < 0.05. **p < 0.01. ***p < 0.001. All communication variables are based on standardized communication scores from the PCA (M = 0, SD = 1). For models predicting T2 aggression perpetration, n = 198; for models predicting T2 relationship satisfaction, n = 144 individuals.

Table 4.

Individual’s Communication Behaviors and Subsequent Relationship Dissolution

		T2 Relationship Dissolution (n = 100 couples)
	Effects Added to Model	B	SE
Hostile	Intercept	−1.76**	.60
	T1 Relationship Satisfaction	−.39*	.17
	Relationship Length	−.22	.19
	Gender	.25	.37
	Hostile Communication	.31	.18
Withdrawn	Intercept	−1.65**	.61
	T1 Relationship Satisfaction	−.35*	.17
	Relationship Length	−.17	.20
	Gender	.15	.38
	Withdrawn Communication	.61**	.20
Vulnerable	Intercept	−1.69**	.60
	T1 Relationship Satisfaction	−.38*	.17
	Relationship Length	−.21	.20
	Gender	.21	.37
	Vulnerable Communication	.09	.17
Warm	Intercept	−1.89**	.61
	T1 Relationship Satisfaction	−.39*	.17
	Relationship Length	−.24	.19
	Gender	.30	.38
	Warm Communication	−.65**	.24
Playful	Intercept	−1.61**	.61
	T1 Relationship Satisfaction	−.40*	.17
	Relationship Length	−.24	.20
	Gender	.12	.38
	Playful Communication	−.56*	.22

Note. *p < 0.05. **p < 0.01. All communication variables are based on individual, standardized communication scores from the PCA (M = 0, SD = 1). The models presented depict results testing one’s own communication as the predictor. As dissolution is a couple-level outcome, all significant findings remain when testing the partner’s communication behavior as the predictor.

Discussion

The present study builds on the idea that the small moments in couples’ lives can have cumulative effects on current and future relationship functioning (Gottman & Driver, 2005; Reblin et al., 2018). Aim 1 descriptive data supported the general acceptability and the feasibility of collecting audio data from couples across various activities and locations throughout the day, and showed that information about context, affect, and verbal content can be meaningfully extracted in 3-minute segments. Aim 2 added nuance to the frequently used ‘positive’ and ‘negative’ categories by distilling 27 microanalytic communication codes into four behavioral constructs of hostility, vulnerability, warmth, and playfulness, plus withdrawal as a stand-alone construct. One’s own and the partner’s hostility, withdrawal, and vulnerability were all associated with T1 aggression, and vulnerability additionally was linked to lower concurrent relationship satisfaction. Consistent with Aim 3 hypotheses, both one’s own and the partner’s hostility were associated with higher aggression and lower relationship satisfaction approximately one year later. Withdrawal showed nuanced effects, with one’s own withdrawal associated with later aggression perpetration, but the partner’s withdrawal was inversely linked to subsequent relationship satisfaction. One’s own playfulness was also associated with lower satisfaction. Warmth and playfulness offered protection against relationship dissolution, whereas withdrawal was associated with an increased likelihood of relationship dissolution.

According to self-reports, participants generally went about their normal routines and engaged in typical interactions during the day of data collection. This window into couples’ private lives depicted a varied range of activities and locations, such as singing together in the kitchen while cooking, or bickering on a street corner about a parking meter. The nature of sampled interactions suggested that couples acclimated to the unobtrusive recordings and that self-presentation bias was minimized. Similar to other naturalistic couple data, (e.g., sample of 10 couples in Alberts et al., 2005), many recordings captured talk about mundane, unremarkable topics, or referred to instrumental tasks (e.g., household chores) or joint activities (e.g., watching TV). Though rarely studied, such habitual, everyday moments make up the bulk of couple interactions and are theorized to be pillars of relationship maintenance (Dainton, 2003).

Our data also show sufficient variability across couples on variables of interest–even within just one day of data collection. We found, for example, that almost all individuals (97.2%) showed at least some degree of hostile behavior, despite a fairly low prevalence in this sample of overt conflict (3.8% of all files; range across couples: 0–32%), on par with similar data collection that coded ~1% of couple interactions as conflictual (Alberts et al., 2005). On the opposite end of the continuum, playful (100% of individuals) and warm (88.7% of individuals) behaviors were widely exhibited. It is, of course, well-known that relationships comprise both negative and positive dimensions, with speculation about the ratio of positive-to-negative behavior needed to keep relationships strong (Gottman et al., 1998). Data here normalize negative interactions, as well as emphasize the importance of positive behaviors for relationship maintenance.

T1 relationship aggression is associated with an increased likelihood of hostile and withdrawing behaviors in couples’ daily interactions. As contrasted with common conceptions of couple aggression as discrete or isolated events, these data indicate that aggression perpetration spills over into hostile behaviors (e.g., critical, rejecting comments and jabs) that emerge in daily conversation. Not surprisingly, past aggression is also associated with both partners’ withdrawal, which may reflect efforts to avoid further confrontations (Burman et al., 1993). Expressions of vulnerability from either partner, typically considered key to increased intimacy, were unexpectedly associated with lower concurrent relationship satisfaction and higher past year aggression. This finding could be a function of the range of behaviors that coalesced in the PCA. However, it also highlights a caution that interactions we strive for in therapeutic contexts may not reflect what serves couples best in all circumstances. Vulnerability may facilitate discussion of relationship problems, particularly with a therapist’s guidance, but may detract from intimacy and closeness if vulnerable disclosures are interpreted by the partner as complaining or implying criticism (Khalifian & Barry, 2020), and thus may be ineffective in daily life.

Our strongest evidence for the relevance of everyday behaviors is their implication for later relationship outcomes. As noted by Kanter et al. (2022), even small effects between communication and later relationship functioning are “striking.” Both partners’ hostility and one’s own withdrawal appear to show an iterative pattern with relationship aggression. That is, past year aggression is associated with hostile and withdrawing behaviors and, in turn, those behaviors also set the stage for subsequent relationship aggression—even controlling for past year aggression. These same behaviors, one’s own hostility and the partner’s hostility and withdrawal, seemingly take a toll on future relationship satisfaction, even with prior satisfaction accounted for. Unexpectedly, one’s own playfulness had an inverse association with subsequent relationship satisfaction. As previously suggested, (Karney & Bradbury, 2020; McNulty, 2010), positive behaviors do not necessarily guard against declines in relationship satisfaction, play a lesser role in portending relationship satisfaction, and, in some cases, may even be deleterious.

For relationship dissolution, however, the picture is more complex, portraying the relevance of both negative (withdrawal) and positive (warmth and playfulness) behaviors. With prior understanding of relationship dissolution coming from lab discussions, these results address the important question of what couples do in their everyday lives that contributes to or detracts from relationship stability. In support of Gottman and Levenson (2002), withdrawal is a risk for relationship dissolution, perhaps by leading to unresolved problems and increased isolation. However, more direct expressions of hostility do not increase risk for dissolution, echoing growing recognition that negative communications are not always detrimental to relationships (Markman et al., 2010; McNulty & Russell, 2010; Overall & McNulty, 2017). Warmth and playfulness appear to promote relationship stability, underscoring the value of small, light-hearted, and kind gestures in couples’ lives. Although our data offer new perspectives on relationship dissolution, additional rigorous tests of these findings, (e.g., through cross-validation; Heyman & Slep, 2001), is warranted. Our results also underscore the importance of context when understanding certain behaviors, such as playfulness. On the one hand, playfulness may support relationship maintenance, fostering enjoyment and connection (De Koning & Weiss, 2002). Alternatively, playfulness can be mean-spirited, distancing, or used to avoid serious topics, potentially damaging relationship quality over time (Hall, 2017). We considered whether playfulness might vary depending upon how well the partners know each other and examined its associations with relationship length and cohabitation status, both of which were nonsignificant. In our view, there is still much to be understood about playfulness, and further exploration of playfulness is best assessed outside of research labs and therapy rooms.

Although substantial literature focuses on newlywed or married couples, interactional patterns can develop sooner, even prior to engagement (Lavner et al., 2012), and endure with relative stability. Negative couple interactions pose a particular threat to health and well-being in emerging adulthood (Gomez-Lopez et al., 2019), especially given that this developmental phase is marked by greatest risk for dating aggression (O’Leary, 1999). As our results suggest, maladaptive communication (i.e., withdrawal, hostility) is a predictor of relationship dysfunction that can be identified quite early in a relationship. Although many romantic relationships in emerging adulthood are not long-lasting, addressing maladaptive interaction processes can have important impacts on current and future romantic relationships. Findings from studying couples in emerging adulthood can have implications for early interventions that can target dysfunctional processes before they solidify or become more resistant to change.

Limitations

Findings should be interpreted in consideration of study limitations. First, although these data represent a much longer sampling of couple behavior than brief lab-based discussions, the data here are based on only one day and reflect an assumption that the sampled interactions are representative of typical patterns. Findings would be bolstered by sampling multiple days of the week (e.g., weekdays versus weekends). Second, we only assessed communication at one time point, which precludes testing how satisfaction influences later communication, the direction accounting for more significant effects in the Lavner et al. (2016) cross-lagged study. Third, an inherent limitation of audio-only data is the loss of nonverbal cues such as facial expressions, gestures, and physical touch, which can convey both affection and intimacy, as well as hostility and aggression. Recent technologies, such as inconspicuously placed motion-detection cameras or “smart-home” devices, may provide solutions to capturing video data (Nelson & Allen, 2018). Fourth, our results collapsed data across the entire day; future work can address how specific behaviors relate to momentary shifts in relationship satisfaction or feelings of irritation with the partner. A more granular analysis of the coded data could investigate time-linked sequences between the two partners, thereby characterizing momentary and directional effects in interaction patterns. Fifth, although we train human coders to make coding decisions based on objective cues and coding guidelines, they still bring their own implicit biases, guided inevitably by cultural and contextual frames of reference (Bulling et al., 2023). We prioritized recruitment of a diverse coding team (approximately 34% Asian/Asian American, 6% Black/African American, 6% Hispanic/Latinx, 15% multiracial, 35% White), but we recognize that coding still may be hampered by lack of familiarity with certain demographic groups or limited understanding of cultural norms, especially in a diverse sample such as this one.

Finally, generalizability of these findings is limited to different-gender, young adult couples. As with other studies on the prediction of relationship outcomes, young couples are often a preferred group to study because communication patterns identified prior to marriage hold considerable relevance for relationship well-being after marriage (Karney & Bradbury, 2020; Lavner et al., 2012; Markman et al., 2010). Nonetheless, we cannot generalize the findings to couples who are older, more established, highly distressed, raising children, and/or same-sex or gender non-conforming. In addition, younger couples, for whom technology is a constant in everyday life, may have been more comfortable with the recording of their daily interactions.

Implications and Conclusions

Findings from this study show that naturalistic, everyday communications between different-gender romantic partners are meaningful for couples’ relationship well-being. In other words, how partners treat each other, even in their seemingly minor interactions, can have significant and long-range consequences. Still to be determined is whether the everyday behaviors identified here are moderated by other personal (e.g., family history or cultural background) and relationship (e.g., attachment styles) characteristics. Moreover, these communications are inherently intertwined with other relationship strengths (e.g., shared values), difficulties (e.g., infidelities), and external stressors (e.g., financial problems).

Capturing naturalistic data that characterizes daily activities and interactions historically has presented undeniable challenges. However, with advances in wearable technologies that can monitor real-time multimodal data in daily life (e.g., including physiology and emotional arousal), as well as algorithms developed through machine learning to extract meaning from these data, it is increasingly feasible to identify couple dynamics of interest and further advance the study of romantic relationships (Timmons, et al., 2017). Although still a preliminary step, the findings here illustrate the relevance of couples’ everyday moments and lend credibility to focusing on daily interactional patterns for couple assessment and intervention. Greater attention to couples’ naturalistic communications can enhance understanding of relationship distress, expand the scope and add precision to communications addressed, and make interventions more translatable to the real world.