443
Views
0
CrossRef citations to date
0
Altmetric
Research Article

How are Conversations via an On-Demand Peer-To-Peer Emotional Well-Being App Associated with Emotional Improvement?

ORCID Icon, , , , &

ABSTRACT

Non-clinical, on-demand peer-to-peer (PtP) support apps have become increasingly popular over the past several years. Although not as pervasive as general self-help apps, these PtP support apps are usually free and instantly connect individuals through live texting with a non-clinical volunteer who has been minimally trained to listen and offer support. To date, there is little empirical work that examines whether and how using an on-demand PtP support app improves emotional well-being. Applying regression and multilevel models to N = 1000+ PtP conversations, this study examined whether individuals experience emotional improvement following a conversation on a PtP support app (HearMe) and whether dyadic characteristics of the conversation – specifically, verbal and emotional synchrony – are associated with individuals’ emotional improvement. We found that individuals reported emotional improvement following a conversation on the PtP support app and that verbal (but not emotional) synchrony was associated with the extent of individuals’ emotional improvement. Our results suggest that online PtP support apps are a viable source of help. We discuss cautions and considerations when applying our findings to enhance the delivery of support provision on PtP apps.

Recent advances in and accessibility of technologies that connect people provide opportunities to deliver social support at scale and at low cost. Recognizing this opportunity, many companies (e.g., Confidist, HearMe, Now&Me, Omegle, 7 Cups) have created on-demand peer-to-peer (PtP) applications (apps) to facilitate conversations between individuals seeking support and non-clinical and minimally trained peers who can positively impact individuals’ emotional well-being. These apps have quickly become a popular source of social support (Clay, Citation2021; Naslund et al., Citation2016). The COVID-19 pandemic further accelerated the use of these apps: Amidst a rise in mental health issues caused by long-term social isolation, loss of family members, and/or job loss, people have been flocking to online apps to connect with others (Daly et al., Citation2022; Lee et al., Citation2020; Wind et al., Citation2020). Given the increased use of these PtP apps for support, new, publicly available research is needed to evaluate their efficacy and to understand the characteristics of app-based PtP conversations, particularly characteristics of dyadic exchange, that may enhance the delivery of support.

Although the efficacy of PtP support apps is understudied, a rich foundation of knowledge on online support groups and in-person PtP support offers evidence of the characteristics of dyadic exchange likely driving improved emotional well-being. Current work on online social support resources, including support groups, forums, and networks where people post and respond to messages, shows that online connections can improve emotional states and provide people with a sense of belonging (Barak et al., Citation2008). From research on in-person PtP support we also know what characteristics of individual messages (Jones et al., Citation2018; MacGeorge & Zhou, Citation2021) and aspects of dyadic exchange contribute to support seekers’ emotional improvement (Burgoon et al., Citation2014; Butler, Citation2011). The features of dyadic interaction are especially relevant to understanding the outcomes of conversations enabled by on-demand PtP support apps. Whereas a support seeker in an online forum or support group may receive a single (or several) message(s) after posting about their stressor and thus might evaluate messages as single instances of attempted support, online PtP support unfolds in ways more akin to a conversation. Thus, our examination of app-based supportive conversations emphasizes dyadic qualities of the communication that occurs on these PtP apps.

Given the rise of PtP support apps as a resource for individuals’ mental health and well-being and the dearth of empirical studies to test the efficacy of non-clinical PtP support apps, we examine whether and how PtP online social support helps individuals improve their emotional state. Specifically, we analyze 1,010 conversations individuals had on the PtP support app HearMe. To the extent that conversations enabled by on-demand PtP support apps lead to emotional improvement in response to daily stressors, they may be useful, low cost-of-entry resources that can assist with rising mental health concerns, especially during extended periods of sustained isolation.

Online social support

Since the arrival of the internet, people have sought the help of others through digital, group-based, self-help networks (Dutta & Feng, Citation2007) and, more recently, digital, one-on-one peer support apps where people chat with a non-clinically trained volunteer peer. Both group-based and peer support apps operate on the principle that support from peers who share or can speak to similar experiences is beneficial to coping and well-being (Braithwaite et al., Citation1999; Wright, Citation2000).

Although researchers have identified many benefits of online social support groups, such as increased access to psychosocial support (Davison et al., Citation2000; Swartwood et al., Citation2011), some characteristics of online social support groups may hinder a sense of connection that contributes to effective support (van Uden Kraan et al., Citation2008). First, the asynchronous nature of online social support groups makes it difficult for individuals to receive immediate help, at the moment they are seeking it. Second, online social support groups are one-to-many forums, meaning an individual’s post is shared with many people. A third issue concerns privacy; while these groups often afford anonymity, individuals may still not feel comfortable discussing their issues in an online space shared with many others. Although online groups provide access to the potential support from a large group of people, it is also possible that no one will respond, that one or more people will respond negatively, or that it will be difficult to establish and maintain a meaningful dialogue. Despite the benefits of online support groups, these are several notable limitations.

PtP support apps address some of the limitations of online social support groups. First, PtP apps connect individuals in real-time. People who use these apps are paired with a listener within a few minutes. Second, these apps recruit and train non-clinical listeners in strategies often framed as “active” or “active-empathic” listening. Research shows that people who interact with listeners trained to actively engage in the conversation, defined by behaviors such as paraphrasing, commenting, asking questions, and being nonverbally immediate, report feeling more understood (e.g., Weger et al., Citation2014) and better about their problem (Bodie et al., Citation2015). Given the distinctive nature of real-time, dyadic, and private PtP support conversations compared to online support groups, empirical testing of the effectiveness of these PtP support apps is valuable.

Digital peer support interventions appear to be feasible and acceptable, with strong potential for clinical effectiveness (Fortuna et al., Citation2020); however, the field is in the early stages of development and requires well-powered efficacy and clinical effectiveness trials. One field study found that mothers experiencing postpartum depression (N = 19) who chatted with trained (but non-clinical, volunteer) listeners on the app 7 Cups outside of regular therapist hours reported improvements in postnatal depression in line with results from a treatment-as-usual control group (Baumel et al., Citation2018). This study suggests the efficacy of PtP support apps, but more research is needed. A rich foundation of research on in-person PtP support will further inform the field about what behaviors of the listener and support seeker are associated with emotional improvements for the support seeker.

Characteristics of supportive talk

Although people seek and provide support in close relationships (Goldsmith, Citation2004), brief conversations with strangers can also effectively contribute to emotional improvement (Cannava & Bodie, Citation2017; Lakey & Orehek, Citation2011; Lepore et al., Citation2000). Indeed, weak social ties play an important role in bolstering day-to-day well-being (Sandstrom & Dunn, Citation2014). Prior research has identified many dyadic features of conversation that facilitate support seekers’ emotional improvement, including dyadic synchrony – i.e., the extent to which two people coordinate and synchronize behavioral, verbal, and emotional cues during talk (Burgoon et al., Citation2014; Butler, Citation2011). Most scholars consider interactional synchrony a universal human behavior (Hatfield et al., Citation1994), and synchrony is a key concept across a variety of theories and psychosocial behaviors, including communication accommodation theory (Soliz et al., Citation2022), interpersonal adaptation theory (Burgoon et al., Citation1995), emotional contagion (Hatfield et al., Citation1994), and the therapeutic alliance (Koole & Tschacher, Citation2016). Across these theories, greater synchrony is hypothesized to be an indication of a positive relationship, as well as increased cooperation, empathy, and shared understanding between dyad members (Bavelas et al., Citation1986; Doré & Morris, Citation2018; Ireland et al., Citation2011; Manson et al., Citation2013; although it may depend on the relational context, cf. Butler, Citation2015). Greater synchrony has also been linked to important prosocial outcomes (Hove & Risen, Citation2009; Roberts et al., Citation2013), affinity and liking (Chartrand & van Baaren, Citation2009), and rapport (Tickle-Degnen & Rosenthal, Citation1990). Across these different theoretical perspectives, synchrony has been studied primarily in a laboratory context (see Mehl & Pennebaker, Citation2003 for an exception). Given that synchrony is a foundational phenomenon for many theoretical perspectives, it is important to examine how synchrony manifests in naturalistic settings and to determine whether the patterns that arise from “actors-behaving-in-context” align with patterns from interactions with artificial constraints imposed by laboratory experiments.

Synchrony can manifest in both verbal and nonverbal behaviors during conversations. Forms of nonverbal synchrony, including conversational partners’ coordination of postural, gestural, and facial coordination, are known to play an important role in social interaction (Burgoon et al., Citation2014). Extensive literatures highlight the consequential impacts of nonverbal/non-content characteristics of conversations, such as backchanneling (e.g., McCarthy, Citation2003), silence/response times (e.g., Templeton et al., Citation2022), and smooth turn taking (e.g., Koudenburg et al., Citation2017). As a conversation level phenomenon, synchrony manifests in behavioral contingencies (i.e., behaviors are proximate causes) of both brief and fast nonverbal behaviors (e.g., Condon & Ogston, Citation1967), in turn-by-turn speaking dynamics (Acosta & Ward, Citation2011), in general behavioral matching during both specific time segments within a conversation (Bodie et al., Citation2015) and over the entire conversation (Guerrero et al., Citation2000; for a summary see; Burgoon et al., Citation1995).

In online support networks and PtP support apps, however, most nonverbal information is not accessible. In these text-based contexts, synchrony may instead manifest in how people coordinate aspects of their language, particularly in how they converse – verbal synchrony – and in the emotions they express – emotional synchrony. Indeed, in the absence of nonverbal information, these aspects of synchrony likely become even more relevant and potentially predictive of conversational outcomes.

Verbal synchrony

The similarity of non-content language during talk may be one way in which verbal synchrony manifests throughout a conversation. Verbal synchrony can be examined using language style matching (LSM), which aims to capture unconscious and stylistic markers of language use. LSM measures the similarity in conversational partners’ use of function words, which are frequently used, typically short, and have little meaning outside of the actual context of the interpersonal exchange (e.g., personal pronouns, articles, conjunctions, prepositions, negations, high-frequency adverbs, quantifiers; Gonzales et al., Citation2010; Ireland et al., Citation2011). LSM has been used as a proxy for the extent to which a listener is engaged with and attuned to their conversation partner, particularly in support interactions (Bowen et al., Citation2017; Cannava & Bodie, Citation2017). Furthermore, higher LSM scores (i.e., greater similarity) were associated with higher ratings of support effectiveness in text-based emotional support conversations (Doré & Morris, Citation2018), greater perceptions of support among a sample of health bloggers and their readers (Rains, Citation2016), and emotional improvement after a five minute, in-person supportive conversation (Cannava & Bodie, Citation2017).

Emotional synchrony

In addition to the similarity of non-content markers of speech, the coordination of emotional expression – often referred to as emotional synchrony – may also contribute to the success of a support conversation. Emotional synchrony is based on the principle that emotions are not only intrapersonal but also interpersonal phenomena; that is, people routinely elicit emotions in each other using verbal (e.g., expressions of feelings in talk) and nonverbal (e.g., glances, gestures) markers (Butler, Citation2015). Emotional synchrony has been used to capture shared understanding between interlocutors (Doré & Morris, Citation2018), with shared understanding being a marker of a listener who may be actively processing, reflecting, and validating the support seekers’ feelings and experiences. Emotional synchrony patterns have been assessed in a variety of ways: by asking partners to review a recorded conversation and to report emotional responses at certain time points (Rohrbaugh et al., Citation2009), by training coders to code a conversation for emotion markers (Roberts et al., Citation2013), and by extracting the sentiment of produced text using natural language processing tools (Tausczik & Pennebaker, Citation2010). Emotional synchrony has been linked to stress and affection during in-person interactions (Roberts et al., Citation2013) and to emotional recovery following a text-based support conversation (Doré & Morris, Citation2018).

Leveraging the text-based nature of the conversations on PtP support apps, we use data from 1000+ conversations that individuals had with peers on HearMe, a PtP support platform. From these conversations, we derived measures of verbal and emotional synchrony and examined how they are related to the support seekers’ emotional improvement. We purposively focus on conversation-level measures of verbal and emotional synchrony that do not depend on topical content of conversations. This allows us to compute and compare features of conversations regardless of what support seekers needed to talk about, and thereby assess the efficacy of a peer-to-peer support app in which users are logging in to discuss a wide variety of issues and topics.

The present study

Based on the prior research on online social support groups and in-person PtP support interactions, we pose the following hypotheses:

H1:

PtP online support conversations result in a positive change in affective state from pre- to post-conversation.

H2:

(a) There is evidence of verbal synchrony, as measured by LSM, between support seekers (users) and support providers (listeners) during an online PtP support conversation, and (b) greater verbal synchrony, as measured by LSM, is associated with greater positive change in affective state from pre- to post-conversation.

H3:

(a) There is evidence of emotional synchrony between support seekers (users) and support providers (listeners) during an online PtP support conversation, and (b) greater emotional synchrony is associated with greater positive change in affective state from pre- to post-conversation.

Empirical work that examines the quickly growing realm of informal telehealth, and understanding whether principles of supportive communication developed in different settings generalize, is of critical applied value. This study seeks to fill the gap in current understanding of the characteristics of PtP computer-mediated conversation, particularly verbal and emotional synchrony, that contribute to support seekers’ emotional improvement in a naturalistic context that has not yet been studied in detail.

Method

Participants

Participants (Ndyads = 1,010) were registered users and listeners of HearMe – a commercially available PtP support app that allows users and listeners to converse in anonymous and confidential conversations. All participants were in the United States, self-reported being at least 18 years old, and participated in conversations on the application between January 1 2021 and September 1 2021 for the primary analyses and between September 2 2021 and April 30 2022 for the replication analyses. HearMe was designed to support and maintain anonymity of support seekers. As such, the platform neither requests nor collects demographic information from their users. These data are therefore not reported here. Similarly, demographic information about the listeners was not shared to preserve their privacy.

Prior to interacting with users, listeners completed a brief training through HearMe that involved watching a series of videos, lasting approximately 15 minutes, that covered how to use the app, how to engage in active listening (i.e., engaging in reflective listening, asking open-ended questions, sharing personal experiences, and providing suggestions instead of advice), and how to handle situations that are not part of HearMe’s mission (e.g., for crisis conversations, listeners are instructed to direct users to an appropriate hotline) or other inappropriate situations (e.g., users requesting listeners’ personal information). Following the training, listeners then completed a series of questions that tested their knowledge and needed to score above 70% to pass the training.

The primary analysis sample consisted of 402 conversations that occurred on HearMe between 273 unique users and 126 unique listeners (some listeners and users had more than one conversation during the sampling period). The replication analysis sample consisted of 608 conversations between 213 unique users and 406 unique listeners.

Procedure

To use HearMe, users are required to download the app and create an account. Prior to each conversation, users are asked to select their preferred language, select a topic of conversation, and rate their affective state; they then engage in a conversation. Immediately after the conversation concludes, users provide a post-conversation affective state rating. Of the initial 916 conversations received for analysis, there were n = 900 in which users reported their pre-conversation affective state, n = 409 in which users reported their post-conversation affective state, and n = 402 in which users reported both their pre- and post-conversation affective state (i.e., had complete data). There were no significant differences in pre-conversation affective state between users in the final analysis sample (N = 402) and all users who reported their pre-conversation affective state, t(811.06), − 1.78, p = .08.

Measures

Pre- and post-conversation affective state

Prior to and immediately after each conversation, users responded to the question: “How are you feeling right now?” using a five-option response scale, terrible (coded as 0) to very good (coded as 4). Studies have shown that single item measures for self-rated mental health (Ahmad et al., Citation2014), happiness (Abdel-Khalek, Citation2006), and life satisfaction (Cheung & Lucas, Citation2014) exhibit qualities that support the validity of a single item to measure affective state. Users’ pre-conversation affective state was, on average, 1.77 (SD = 0.87, Median = 2.0, Range = 0.0 to 4.0), and users’ post-conversation affective state was, on average, 2.86 (SD = 0.87, Median = 3.0, Range = 1.0 to 4.0). The percentage of responses for each response option are given in Table S1. Affective state change was calculated as the difference between the post-conversation and pre-conversation affective state reports. Positive values (>0) indicate that a user’s affective state improved.

Conversation-level verbal synchrony

Verbal synchrony is often operationalized using LSM, a measure of the correspondence between the stylistic, non-content characteristics of speech between dyad members (Gonzales et al., Citation2010; Ireland et al., Citation2011). LSM was quantified as the extent of overlap in proportional use of nine different types of function words, identified using the Linguistic Inquiry and Word Count (LIWC) program (i.e., articles, auxiliary verbs, conjunctions, high-frequency adverbs, impersonal pronouns, negations, personal prepositions, pronouns, and quantifiers; Pennebaker et al., Citation2015), that were produced by each partner during the entirety of the conversation.Footnote1 Specifically, LSM was calculated as

(1) LSM=1Kk=1K1π1kπ2kπ1k+π2k+0.0001(1)

where k = 1 to K are the function word categories (in our case K = 9) and π1k and π2k are the proportion of function words in each category for the user and listener, respectively. On average, LSM was 0.83 (SD = 0.07, Median = 0.84, Range = 0.52 to 0.97).

Conversation-level emotional synchrony

Emotional synchrony was operationalized as a conversation-level measure of the correspondence between users’ and listeners’ emotional expression across speaking turns and was invoked in two steps. First, the level of emotional sentiment for each speaking turn, where “speaking” turn is defined as a series of consecutive text messages from each partner, was measured using the most recently released and largest sentiment dictionary (that we are aware of): the NRC Valence, Arousal, and Dominance (VAD) Lexicon, which contains 20,000 English words that were human rated for valence, arousal, and dominance (Hipson & Mohammad, Citation2021; Mohammad, Citation2018). Using this dictionary, the text content of each speaking turn was assigned an emotional valence rating that indicated the negativeness (score of 0) to positiveness (score of 1) of that turn. Across all user turns, the average emotional valence was Muser = 0.65 (SD = 0.14, Median = 0.64, Range = 0.00 to 1.00); and across all listener turns, average emotional valence was Mlistener = 0.65 (SD = 0.13, Median = 0.65, Range = 0.00 to 1.00). Second, as detailed below in the Data Analysis section, the extent of emotional synchrony in each conversation was quantified within a multilevel model that accommodated the nested nature of the data (speaking turns within conversations) as the relation between the emotional sentiment of the user’s turns and the emotional sentiment of the listener’s turns.

Given that different dictionaries can produce different sentiment scores, the robustness of the findings was examined in supplementary analyses using six alternative sentiment analysis tools: AFINN dictionary (Nielsen, Citation2011), Bing dictionary (Hu & Liu, Citation2004), NRC Lexicon (Mohammad & Turney, Citation2010), Syuzhet dictionary (Jockers, Citation2015), LIWC affect dictionary (Pennebaker et al., Citation2015), and the sentimentr package in R (Rinker, Citation2021) which accounts for valence shifters (e.g., not good). Sentiment was calculated using the (a) tidytext package in R (R Core Team, Citation2022) for the NRC-VAD dictionary (Silge & Robinson, Citation2016); (b) syuzhet package in R (R Core Team, Citation2022) for the AFINN, Bing, NRC, and Syuzhet dictionaries (Jockers, Citation2015); (c) LIWC program for the LIWC affect dictionary (Pennebaker et al., Citation2015); and (d) the sentimentr package in R for valence shifters (Rinker, Citation2021). All of the selected sentiment analysis tools used in this study underwent extensive human validation processes by their initial creators, and many of these sentiment analysis tools have been used in similar contexts (e.g., social support in social media posts using LIWC: Warner et al., Citation2018; a version of the Bing dictionary was used to examine emotions and opinions of individuals posting in online breast cancer support groups: Cabling et al., Citation2018). Additional details about each of the sentiment analysis tools can be found in the supplemental material. Descriptive statistics for the six alternative sentiment analysis tools are given in Table S2 of the supplementary material.

Data analysis

Are conversations on a PtP support app associated with emotional improvement?

To examine whether a conversation with a listener on HearMe results in a change in user reported emotional improvement (H1), we fit a linear regression model in which affective state change was regressed on centered pre-conversation affective state. Of interest is whether the intercept, which represents the expected affective state change controlling for pre-conversation differences in affective state, is significantly different than zero.

Is there verbal synchrony between users and listeners, and is the extent of verbal synchrony associated with emotional improvement?

We fit a one sample t-test to examine whether there is evidence of verbal synchrony (H2a). To examine whether differences in verbal synchrony are related to differences in users’ emotional improvement (H2b), we fit a multiple linear regression model in which affective state change was regressed on dyads’ LSM score, while controlling for users’ pre-conversation affective state (centered) and the total number of turns within the conversation.

Is there emotional synchrony between users and listeners, and is the extent of emotional synchrony associated with emotional improvement?

Emotional synchrony was operationalized as a conversation-level measure of the correspondence between users’ and listeners’ emotional expression across speaking turns. The prototypical level of emotional synchrony (H3a) and whether differences in emotional synchrony were associated with differences in users’ emotional improvement (H3b) were examined using a multilevel model in which listeners’ turn-level emotional sentiment was regressed onto the users’ turn-level emotional sentiment (i.e., state sentiment, the deviation from the individuals’ mean-level of emotional sentiment), the users’ overall level of emotional sentiment (i.e., trait sentiment, individuals’ mean-level emotional sentiment), the users’ affective state change, and the cross-level interaction of users’ state emotional sentiment and affective state change,Footnote2 while controlling for the cross-level interactions of users’ state and trait turn-level emotional sentiment and users’ trait-turn level emotional sentiment and affective state change. Specifically,

(2) ListenerSentimentdt=β0d+β1dStateUserSentimentdt+β2dTurnPairdt+edt(2)

where ListenerSentimentdt is the observed listener emotional sentiment for dyad d at turn pair t; β0d is the prototypical level of listener emotional sentiment for dyad d; β1d is the within-dyad association between user and listener sentiment across turns and quantifies the extent of emotional synchrony for dyad d; β2d is the rate at which listener emotional sentiment changes over the course of the conversation (i.e., over time); and edt are turn pair-specific residuals that are assumed to be normally distributed. Dyadic-specific parameters were modeled as

(3) β0d=γ00+γ01TraitUserSentimentd+γ02AffectChanged+γ03TraitUserSentimentdAffectChanged+u0d(3)
(4) β1d=γ10+γ11TraitUserSentimentd+γ12AffectChanged+u1d(4)
(5) β2d=γ20(5)

where γ00 γ01, γ02, γ03 indicate the expected value of listener overall emotional sentiment for the prototypical dyad in the sample, and how differences in listener overall sentiment are related to users’ overall emotional sentiment, users’ affective state change, and their interaction, respectively; where γ10 γ11, γ12 indicate the expected value of emotional synchrony for the prototypical dyad in the sample, and how differences in emotional synchrony are related to users’ overall emotional sentiment and users’ affective state change, respectively; where γ20 indicates how much listeners’ emotional sentiment changes over the course of the conversation (i.e., time); and where u0d and u1d are unexplained between-dyad differences in the expected levels of listener emotional sentiment and emotional synchrony that are assumed to be multivariate normally distributed.

is a graphical representation of the multilevel model used to examine hypotheses H3a and H3b (EquationEquations 2Equation5). For evidence of emotional synchrony (H3a), we examined whether the parameter γ10 is greater than zero. Notably, this parameter describes emotional synchrony at the conversation-level, that is, how much user’s and listener’s emotional sentiment fluctuates together across the entire conversation. For evidence of a relation between emotional improvement and emotional synchrony (H3b), we examined the parameter γ12. These relations of interest are highlighted with bold arrows in .

Figure 1. Graphical Representation of the Model to Examine Emotional Synchrony.

Note. The figure is a graphical representation of the multilevel model used to examine hypotheses H3a and H3b (EquationEquations 2Equation5). The rectangles represent manifest variables, the circles represent latent variables, the two headed arrows represent variances and covariances, the one-headed arrows represent regressions, and where the triangles represent vectors of 1 that invoke intercepts. In this graphical representation of the model that was fit to the observed data, the intercepts (β0d) and slopes (β1d) of the within-dyad relations, where listeners’ time-varying emotion sentiment are regressed on users’ time-varying sentiment, are shown as small circles on their associated arrows (i.e., the random effects). These parameters are then projected and summarized as Overall Listener Emotion Sentiment and Emotional Synchrony factors that are then regressed at the between-dyad level on the Trait User Sentiment and Affect Change variables and their interaction. The relations of interest for testing H3a (γ10, the prototypical level of emotional synchrony) and H3b (γ10 commonly used emojis we12, the relation between differences in affect change and emotional synchrony) are shown as bold lines.
Figure 1. Graphical Representation of the Model to Examine Emotional Synchrony.

We examined the same set of relations using the same model set up with emotional sentiment variables derived from six alternative sentiment tools to assess the robustness of these findings. The multilevel models were fit in R (R Core Team, Citation2022) using the nlme package (Pinheiro et al., Citation2020). Listener and user turn pairs were aligned so that every conversation began with a user turn (i.e., if a conversation began with a listener turn, that turn was removed). Plots were constructed in R using the ggplot2 package (Wickham, Citation2016).

Replication study and analysis

For the replication analyses, we obtained 614 conversations with complete pre- and post-conversation affective state data.Footnote3 There were no significant differences in pre-conversation affective state between users included in the final replication analysis sample and the larger set of users who reported their pre-conversation affective state during the collection period, t(1119.9), − 1.00, p = .32. Notably, the separately obtained primary and replication analysis samples were comparable and did not differ on their average pre-conversation affective state (t(883.49), − 0.12, p = .90), post-conversation affective state (t(894.50), 1.80, p = .07), or affective state change (t(911.15), 1.57, p = .12). All data preparation and analyses followed the procedure described above, with the exception that a small number of conversations (n = 6) that consisted of a single message were removed (this did not occur in the primary sample).

Results

The primary analysis sample included 24,794 messages exchanged during 15,068 speaking turns (i.e., consecutive text messages from one dyad member) within 402 PtP conversations. These conversations lasted, on average, about half an hour (M = 32.13 minutes, SD = 20.99 minutes, Median = 26.97, Range = 4.01 to 151.22 minutes) and consisted of around 60 messages (M = 61.68, SD = 51.44, Median = 45.00, Range = 18.00 to 467.00). On average, a conversation had 37.48 speaking turns (SD = 29.18, Median = 30.50, Range = 9.00 to 306.00), with each turn containing, on average, 24.32 words (SD = 30.03, Median = 16.00, Range = 0.00 to 632.00).Footnote4 Users frequently reported wanting to discuss relationships (n = 134) and “just wanting to talk” (n = 124). Table S3 contains a complete list of conversation topics and frequencies.

Users generally reported an improvement in affective state, with an average affective state change of slightly over one scale point (M = 1.09, SD = 1.05, Median = 1.0, Range = −2.0 to 4.0). depicts users’ change in affective state. Specifically, users’ pre- and post-conversation affective state is plotted along the x-axis, with lines connecting individual user’s reports. Each small dot represents an individual’s report of affective state, with darker and more dense clusters of dots indicating more individuals reporting that level of affective state, and the dots (i.e., affective state reports) being jittered slightly to reduce the overlap of points and facilitate visibility. For instance, the dense cluster of points on the left indicates that most users reported an affective state of “” (or 2) or lower prior to the conversation. Following the conversation, the dense cluster of points on the right is shifted upwards, indicating an improvement in affective state. The lines connecting pre- and post-conversation affective state represent change in affective state, with thicker lines indicating where more individuals followed a particular trajectory (e.g., their affective state rating improving to 3 from 2). As evident by the few downward sloping lines, not all users’ affective state improved.

Figure 2. Change in Users’ Affective State from Pre- to Post Conversation.

Note. Users’ pre- and post-conversation affective state is plotted along the x-axis, with lines connecting individual user’s reports. Each small dot represents an individual’s report of affective state, with the dots (i.e., affective state reports) being jittered slightly to reduce the overlap of points and facilitate visibility. The lines connecting pre- and post-conversation affective state represent change in affective state, with thicker lines indicating where more individuals followed a particular trajectory.
Figure 2. Change in Users’ Affective State from Pre- to Post Conversation.

Are conversations on a PtP support app associated with emotional improvement?

Results from the linear regression testing whether conversations on HearMe are associated with users’ emotional improvement (H1) are presented in . This model explained a significant amount of variance in users’ affective state change, F(1, 400) = 225.20, p < .05, R2 = .36, R2Adjusted = .36. We found that, even after controlling for differences in pre-conversation affective state, affective state change was positive (b = 1.09, SE = 0.04, p < .01). There was, however, evidence of a ceiling effect: The higher users’ affective state was prior to the conversation, the less their affective state improved following the conversation (b = −0.72, SE = 0.05, p < .01). To account for the few dyads that had multiple conversations with one another, we refit the model in a multilevel modeling framework to accommodate the data dependencies and found similar results to those reported from the linear regression model (see Table S4 of the supplemental material). From these results, we conclude that H1 was supported.

Table 1. Results from regression examining change in users’ affective state.

Is there verbal synchrony between users and listeners, and is the extent of verbal synchrony associated with emotional improvement?

First, we found evidence of verbal synchrony between users and listeners. As consistent with H2a, the average level of verbal synchrony was significantly different from zero, t(401) = 235.21, p < .001. Second, the results from the regression model examining the association between LSM and affective state change are presented in . This model explained a significant amount of variance in users’ emotional improvement, F(3, 398) = 88.87, p < .05, R2 = .40, R2Adjusted = .40. As predicted by H2b, there was a significant association between LSM and users’ affective state change (b = 2.55, SE = 0.61, p < .01). As shown in , higher levels of LSM are associated with greater improvements in users’ affective state. Importantly, this association is evident after controlling for differences in pre-conversation affective state (b = −0.73, SE = 0.05, p < .01) and the total number of messages exchanged (b = 0.002, SE = 0.001, p = .11). To account for the few dyads that had multiple conversations with one another, we refit the model in a multilevel modeling framework to accommodate the data dependencies and found similar results to those obtained with the multiple linear regression model (see Table S5 of the supplemental material). From these results, we conclude that H2 was supported.

Figure 3. Results from Regression Model Predicting Affective State Change from Language Style Matching.

Note. The figure depicts the association between language style matching and affective state change (H2b).
Figure 3. Results from Regression Model Predicting Affective State Change from Language Style Matching.

Table 2. Results from regression examining the association between lsm and change in users’ affective state.

Is there emotional synchrony between users and listeners, and is the extent of emotional synchrony associated with emotional improvement?

Results from the multilevel model examining emotional synchrony and its relation to emotional improvement are presented in . We found that the emotional synchrony parameter (γ10) was positive and significantly different from zero. Specifically, the prototypical within-dyad association between listeners’ emotional sentiment and users’ state emotional sentiment was γ10 = 0.12 (SE = 0.02, p < .01), indicating emotional synchrony, such that on turns when users’ emotional sentiment was higher (i.e., more positive) than average, listeners’ emotional sentiment also tended to be higher than average. As we would expect, there were between-dyad differences in both listeners’ expected level of emotional sentiment (u0d) and dyads’ emotional synchrony (u1d). From these results, we conclude that H3a was supported.

Table 3. Results from multilevel model examining the association between users’ and listeners’ emotional sentiment and change in users’ affective state.

There was also a significant association between listeners’ overall level of emotional sentiment and users’ trait emotional sentiment (γ01 = 0.47, SE = 0.06, p < .01), indicating that users’ who, on average, expressed more positivity were conversing with listeners who, on average, expressed more positivity. Although not hypothesized, there was a positive association between turn pair (i.e., time in the conversation) and listener emotional sentiment (γ20 = 0.10, SE = 0.00, p < .05), suggesting that, on average, listeners’ emotional sentiment increased over the conversation. As well, there was a negative association between users’ affective state change and listener emotional sentiment (γ02 = −0.01, SE = 0.00, p < .01).

We also examined whether users’ emotional improvement was associated with emotional synchrony. There was no evidence that users’ emotional improvement (i.e., affective state change) was related to differences in the extent of synchrony between listeners’ sentiment and users’ state sentiment (γ12 = 0.02, SE = 0.02, p = .28). Thus, we conclude that H3b was not supported.

Robustness checks

Given the different dictionaries and utilization of sentiment analysis tools across the field, we checked whether the results obtained using the NRC-VAD dictionary were robust when using other tools: AFINN, Bing, NRC, Syuzhet, LIWC affect, and sentimentr. Results from these six alternative H3 models are displayed in the supplemental material in Table S6. All six tools provided evidence of emotional synchrony (H3a); that is, there was always a significant positive association between state user emotional sentiment and listener emotional sentiment, as consistent with our original analyses. We also found that there was a positive association between trait user emotional sentiment and listener emotional sentiment for five of the six tools (all but LIWC affect, which was non-significant). Across three of the six tools (the AFINN, NRC, and sentimentr tools), and consistent with H3b (but inconsistent with our findings from the NRC-VAD dictionary), we found that users who reported greater affective state change had higher emotional synchrony. Thus, the collection of results suggests that the findings for H3a are generally robust across sentiment analysis tools, whereas the findings for H3b are inconsistent across sentiment analysis tools.

Replication study results

The replication analysis sample included 44,287 messages exchanged within 608 PtP conversations. On average, conversations took place over the course of half an hour (M = 31.32 minutes, SD = 38.66, Median = 23.82, Range = 0.40 to 714.00), consisted of around 70 messages (M = 72.84, SD = 71.04, Median = 55.00, Range = 2.00 to 954.00) and around 40 turns (M = 43.25, SD = 42.23, Median = 32.50, Range = 2.00 to 628.00). When initiating these conversations, users frequently reported “just wanting to talk” (n = 310) and wanting to discuss relationships (n = 237). Table S3 in the supplemental material contains a complete list of conversation topics and frequencies.

Results from the replication sample corroborated the findings obtained in the primary sample. In support of H1, and consistent with the primary analyses, users experienced emotional improvement (regression results in ). Similarly, in support of H2 and consistent with the primary analyses, higher verbal synchrony as measured by LSM was associated with greater emotional improvement (regression results in ). Finally, in support of H3a and consistent with the primary analyses, there was evidence of emotional synchrony between users and listeners. Also consistent with the primary analyses, there was no evidence that differences in users’ affective state change was related to differences in the strength of emotional synchrony (multilevel model results in Tables S7 and S8 of the supplemental material). In sum, the same pattern of findings obtained in the primary analyses emerged in the replication analyses.

Discussion

In this paper, we examined whether conversations that occur on a PtP support app were associated with improvements in users’ emotional well-being and what aspects of the dyadic exchange – specifically, verbal and emotional synchrony – were particularly associated with greater user emotional improvement. We found that, on average, users’ affective state increased following the conversation with a HearMe listener. Higher levels of verbal synchrony in the conversations, as measured by LSM, were associated with greater user emotional improvement. Furthermore, there was emotional synchrony in the turn-by-turn emotional sentiment expressed by users and listeners during the conversations, but there was little to no evidence that the strength of emotional synchrony was associated with differences in users’ reported emotional improvement. Overall, the findings of this study inform our understanding of the characteristics of computer-mediated conversation that contribute to support seekers’ emotional improvement in an online context (PtP support apps) that individuals are increasingly using as a resource for help with everyday stressors.

Theories that capture synchrony, such as interaction adaptation theory (Burgoon et al., Citation1995) or communication accommodation theory (Giles et al., Citation1991), predict that people who match language are more favorably viewed with respect to perceived warmth and liking. Empirical evidence tends to confirm these predictions (Gasiorek, Citation2016). The current study makes an important contribution to these theories by extending findings to naturalistic online peer-to-peer interactions, thus expanding our understanding of the boundary conditions under which synchrony and its impacts occur. This is an important point because the majority of synchrony research examines synchrony in the lab. In this study, while all listeners were trained in basic active listening skills, the conversations were unconstrained, unscripted, and unstructured, thus capturing users’ naturally occurring online conversational behaviors. Examining how motivated individuals seek and provide support in real-life, rather than in artificial conversations among (often college student) participants, allows us to test whether synchrony and its effects are robust in a more naturalistic context.

Furthermore, research that has examined synchrony in naturalistic online interactions (e.g., Rains, Citation2016) has generally done so in one-to-many settings (although see Doré & Morris, Citation2018 for an exception). Our study extends this prior work by focusing on one-to-one interactions, which is especially important given the need to assess the efficacy of mental health apps and whether our knowledge of synchrony in other settings (e.g., dyadic interactions in the lab) is applicable in this context (Clay, Citation2021). Although we do not have observations of mechanisms that might explain how or why synchrony is (not) associated with emotional improvement in novel ways, our study shows that even in relatively short, zero-history, online interactions, people match behaviors, which, in turn, influence emotion regulation, albeit to a small degree.

Benefit of online PtP support

Users of the HearMe PtP support app reported emotional improvement following a single conversation with a listener. This finding is consistent with past work demonstrating that conversations with a stranger who is not a mental health professional can be beneficial (Cannava & Bodie, Citation2017). Past work also shows, however, that not all ways of interacting with others are equally beneficial. When support providers invalidate felt emotions, for instance, the benefits of conversation are diluted (e.g., Lepore et al., Citation2000).

Indeed, the results of this study are somewhat consistent with current understanding of the characteristics of dyadic exchange that contribute to support seekers’ emotional improvement. On one hand, our finding that verbal synchrony was positively associated with emotional improvement is consistent with prior work examining verbal synchrony in perceptions of support online (Rains, Citation2016), in online support conversations (Doré & Morris, Citation2018), and in in-person support interaction (Cannava & Bodie, Citation2017). On the other hand, we did not find evidence that the extent of emotional improvement was associated with emotional synchrony. The inconsistency with both our hypothesis and prior work (e.g., Doré & Morris, Citation2018) may be a result of several issues. First, our study differs from prior work on PtP support (Doré & Morris, Citation2018) that assessed emotional improvement at a greater time lag (30 and 60 minutes following the conversation) and examined nonlinearities in emotional improvement. Second, helpful statements from the support provider, such as validating the support seeker’s expressed emotions (i.e., high person-centered messages, High & Dillard, Citation2012), may not contribute to emotional synchrony as we have operationalized it. Similarly, the ways in which support providers can help support seekers reassess their problem in a less distressing manner may not entail strong alignment with the support seekers’ expressed emotions. Third, interpersonal synchrony is a complex phenomenon that is defined and measured in a variety of ways and at different levels (Delaherche et al., Citation2012), and each of these may impact emotional improvement in different ways across contexts. Keeping in mind these complexities, our findings suggest that verbal and emotional synchrony do surface in text-based support conversations, and that differences in verbal synchrony may contribute to emotional improvement.

Considerations for online PtP support

This paper contributes to the much-needed study of online mental health and wellness apps. Despite our optimism about online PtP support apps and their ability to provide care and connection in new ways, we want to temper readers’ potential interpretation of our findings.

First, we found that higher levels of verbal synchrony, as measured by LSM, was associated with greater emotional improvement. We caution readers and app developers from encouraging listeners to mimic users’ language as it risks overaccommodation. Prior work has shown that seemingly forced similarity can be perceived as patronizing (Dragojevic et al., Citation2016), which could lead to support seekers feeling worse about themselves or their stressor. Furthermore, LSM is considered an “unconscious” marker of language use; so presumably it is difficult to purposively change one’s use of function words to match another’s. Future research can explore how listener training influences provision of quality support (e.g., reflective listening) that may also have concurrent and positive impacts on synchrony.

Second, we encourage readers to use caution when applying our findings to severe problems. Most PtP support research, including this study, has focused on everyday stressors. Our findings suggest that online PtP support can be a viable care option for individuals who experience mild and/or subclinical mental health symptoms. Online PtP support is accessible, given its low cost to users and the high prevalence of smartphone use across demographic groups in the United States (Pew Research Center, Citation2021), and can provide an alternative path to receive help, particularly for those whose symptoms may not necessitate traditional mental health care. The connections that online PtP can foster may help combat the high rate of loneliness in the United States, which, in turn, is associated with a variety of serious health issues such as depression and even mortality (Holt-Lunstad, Citation2017; Holt-Lunstad et al., Citation2015). Treatment for more severe or clinical issues, however, are unlikely to be resolved by simply engaging in supportive talk online with an untrained professional. Even so, technology enabled supportive conversations may be an effective adjunctive option to assist recovery for those already receiving professional support, especially given the lack of support for using stand-alone apps to treat severe mental health conditions (Weisel et al., Citation2019). We encourage researchers to conduct additional work on whether and how online support and treatment can be provided through apps for clinical issues (alone or as part of existing treatment plans) given the advantages these online resources may provide (e.g., increased accessibility, lower costs of treatment).

Limitations and future directions

Although we are enthusiastic about the potential positive impacts that we found in one PtP support app, these findings should be interpreted in light of a few limitations. First, there are limitations with respect to the individuals in our sample. HearMe users are a self-selected group of individuals who willingly chose to seek support from a stranger online. These individuals may be a set of people who are comfortable in this context and/or are willing to seek help. Furthermore, we know little about the users’ or listeners’ demographics (e.g., age, gender, race/ethnicity), clinical history, and training (in the case of listeners who may have experience in helping professions). Future work should collect additional information about individual demographic and clinical characteristics or experiences that may impact the use and success of PtP support apps and explore whether individuals who are encouraged or assigned to conversations on the app experience similar benefits as individuals who sought out this experience themselves.

Second, our results may be specific to the measures of emotional improvement and conversation characteristics used in our analysis. Users’ affective state was measured prior to and following the conversation using one relatively blunt (i.e., a 5-point) item. Future work should explore whether and how an online conversation impacts certain emotions or experiences, such as sadness, stress, or loneliness, to better understand what mental health and wellness problems these types of apps can address. Furthermore, the in-app measurement of emotional improvement may be prone to testing effects – i.e., that participants may feel they are supposed to indicate improved affective state following the conversation. Future work can examine app users over multiple occasions to better assess the extent of testing effects and better isolate systematic changes that follow conversations. Similarly, we measured characteristics of the dyadic exchange using LSM and emotional sentiment metrics that relied on computational approaches that may miss more nuanced aspects of conversations (e.g., reflection, advice provision) and may be somewhat unreliable for especially short texts (Eichstaedt et al., Citation2021). As well, many of the tools we used to measure sentiment used a unigram (i.e., single word) approach that ignored qualifying words (except for the sentimentr tool; Rinker, Citation2021) and did not incorporate emotional information conveyed using emojis. Generalization from our specific operationalization of verbal and emotional synchrony should be done cautiously. Finally, the observational nature of the data and the structure of the multilevel model constrained both the ways the implied temporal sequencing of changes could be operationalized and the opportunity to establish causality. We look forward to future work that (a) utilizes novel computational approaches to capture the complexity of language, such as tools that integrate text and emojis; (b) incorporates human-assessed qualities of support provision and the conversation generally; (c) includes different metrics and approaches to capture conversation dynamics (e.g., the timing of the listener asking questions or providing advice); and (d) structured interventions that support causal inference about how conversation dynamics impact PtP app users’ emotional improvement.

Third, our sample included a small subset of dyads that had multiple conversations with each other. Although our supplementary analyses obtained the same pattern of results when taking the nesting into account, we might expect differences in the qualities and impacts of conversations in which dyad members have developed a relationship. Future work can explore the difference between conversations and their impact in zero-history online relationships versus relationships that involve prior interactions and can also leverage experimental designs with carefully constructed control groups (or conversations) and causal inference-oriented differences-in-differences modeling approaches.

Lastly, we are sensitive to the vast and complex work on interpersonal synchrony. There are numerous temporal response patterns and dynamics that can be used to characterize dyadic synchrony (Burgoon et al., Citation1995). Our data captured synchrony in only two ways. We envision more nuanced analyses as a potential next step, such as capturing specific response contingencies (i.e., Partner A’s behavior is reciprocated with the same behavior from Partner B), incorporating information about the specific content (e.g., advice) of the messages, and examining the valence of synchrony (i.e., whether synchrony of positive versus negative emotions has differential effects).

Conclusion

PtP social support online seems to be a viable source of support for everyday issues and can help address prevalent public health issues, such as loneliness (Holt-Lunstad, Citation2017; Holt-Lunstad et al., Citation2015). This study provides some of the first steps to understand whether PtP support apps used in everyday life can help with users’ emotional improvement and what characteristics of the dyadic exchange may contribute to that emotional improvement. We began by examining verbal and emotional synchrony as particular characteristics of dyadic exchange that may help users. As future work extends into the many other aspects of conversations worth additional study, we look forward to further understanding dyadic support conversations and their impacts, particularly in online spaces.

Supplemental material

OnlinePtPSupport_supplementaryMaterials.docx

Download MS Word (43.4 KB)

Acknowledgements

Thank you to HearMe for sharing their data and making this project possible. The content is solely the responsibility of the authors and does not necessarily represent the official views of HearMe.

Disclosure statement

Michael L. Birnbaum provided consulting services to HearMe and received salary support. The remaining authors report there are no competing interests to declare.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/10410236.2024.2360178.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.

Notes

1. Misspellings in the conversations were not corrected. We assume that misspellings were randomly distributed across conversations and function words.

2. Because change in affective state from pre- to post-conversation is the time-invariant variable (i.e., it manifests between-dyads), it can only be modeled as a moderator (i.e., interaction) of the within-dyad association (i.e., the association between user and listener sentiment), rather than as the outcome of turn-to-turn experiences. That is, multilevel modeling requires the most granular, time-varying level of measurement (in this case, the turn-level) to be the outcome modeled at Level 1 of the multilevel model (EquationEquation 2), and the more global, time-invariant level (in this case, change in affective state) to be a predictor/moderator modeled at Level 2 of the multilevel model (EquationEquations 3Equation5). Because we are in a correlational framework, the hypothesis is articulated by the interaction term that specifically examines whether between-dyad differences in affective state change are related to synchrony, which is captured by the within-dyad association between users’ and listeners’ sentiment. We recognize this approach is atypical, but prior work examining changes in relational characteristics (e.g., Brisini et al., Citation2023) and changes in health (e.g., Schöllgen et al., Citation2016) over time have taken similar approaches.

3. Of the 1,791 conversations received for the replication analysis, there were n = 1,744 conversations in which users reported their pre-conversation affective state, n = 616 conversations in which users reported their post-conversation affective state, and n = 614 conversations in which users reported their pre- and post-conversation affective state.

4. Emojis were not counted as words; hence, a turn could have zero words. To better understand the impact of excluding emojis in our analysis, we examined the number of emojis, the number of speaking turns that had emojis, and the most frequently used emojis in the messages sent in our primary analyses. We found (a) a total of 1,301 emojis were used (<1% out of a total of 366,472 words), (b) 905 speaking turns contained emojis (~6% out of a total of 15,068 speaking turns), and (c) the most commonly used emojis were (n = 252), (n = 77), (n = 74), and (n = 56). Given their relatively low frequency and the potential ambiguity of some of the emojis (e.g., ), we did not include them in the analyses here.

References

  • Abdel-Khalek, A. M. (2006). Measuring happiness with a single-item scale. Social Behavior & Personality: An International Journal, 34(2), 139–150. https://doi.org/10.2224/sbp.2006.34.2.139
  • Acosta, J. C., & Ward, N. G. (2011). Achieving rapport with turn-by-turn, user-responsive emotional coloring. Speech Communication, 53(9–10), 1137–1148. https://doi.org/10.1016/j.specom.2010.11.006
  • Ahmad, F., Jhajj, A. K., Stewart, D. E., Burghardt, M., & Bierman, A. S. (2014). Single item measures of self-rated mental health: A scoping review. BMC Health Services Research, 14(1), 1–11. https://doi.org/10.1186/1472-6963-14-398
  • Barak, A., Boniel-Nissim, M., & Suler, J. (2008). Fostering empowerment in online support groups. Computers in Human Behavior, 24(5), 1867–1883. https://doi.org/10.1016/j.chb.2008.02.004
  • Baumel, A., Tinkelman, A., Mathur, N., & Kane, J. M. (2018). Digital peer-support platform (7Cups) as an adjunct treatment for women with postpartum depression: Feasibility, acceptability, and preliminary efficacy study. Journal of Medical Internet Research MHealth and UHealth, 6(2), e38. https://doi.org/10.2196/mhealth.9482
  • Bavelas, J. B., Black, A., Lemery, C. R., & Mullett, J. (1986). “I show you how I feel”: Motor mimicry as a communicative act. Journal of Personality and Social Psychology, 50(2), 322–329. https://doi.org/10.1037/0022-3514.50.2.322
  • Bodie, G. D., Vickery, A. J., Cannava, K., & Jones, S. M. (2015). The role of “active listening” in informal helping conversations: Impact on perceptions of listener helpfulness, sensitivity, and supportiveness and discloser emotional improvement. Western Journal of Communication, 79(2), 151–173. https://doi.org/10.1080/10570314.2014.943429
  • Bowen, J. D., Winczewski, L. A., & Collins, N. L. (2017). Language style matching in romantic partners’ conflict and support interactions. Journal of Language and Social Psychology, 36(3), 263–286. https://doi.org/10.1177/0261927x16666308
  • Braithwaite, D. O., Waldron, V. R., & Finn, J. (1999). Communication of social support in computer-mediated groups for people with disabilities. Health Communication, 11(2), 123–151. https://doi.org/10.1207/s15327027hc1102_2
  • Brisini, K. S., Solomon, D. H., & Brinberg, M. (2023). Linking stress to the emergence of relational turbulence in marriage. Personal Relationships, 30(4), 1383–1404. https://doi.org/10.1111/pere.12515
  • Burgoon, J. K., Dunbar, N. E., & White, C. (2014). Interpersonal adaptation. In C. R. Berger (Ed.), Interpersonal communication (pp. 225–248). De Gruyter Mouton Publisher.
  • Burgoon, J. K., Stern, L. A., & Dillman, L. (1995). Interpersonal adaptation: Dyadic interaction patterns. Cambridge University Press.
  • Butler, E. A. (2011). Temporal interpersonal emotion systems: The “TIES” that form relationships. Personality and Social Psychology Review, 15(4), 367–393. https://doi.org/10.1177/1088868311411164
  • Butler, E. A. (2015). Interpersonal affect dynamics: It takes two (and time) to tango. Emotion Review, 7(4), 336–341. https://doi.org/10.1177/1754073915590622
  • Cabling, M. L., Turner, J. W., Hurtado de Mendoza, A., Zhang, Y., Jiang, X., Drago, F., & Sheppard, V. B. (2018). Sentiment analysis of an online breast cancer support group: Communicating about tamoxifen. Health Communication, 33(9), 1158–1165. https://doi.org/10.1080/10410236.2017.1339370
  • Cannava, K., & Bodie, G. D. (2017). Language use and style matching in supportive conversations between strangers and friends. Journal of Social and Personal Relationships, 34(4), 467–485. https://doi.org/10.1177/0265407516641222
  • Chartrand, T. L., & van Baaren, R. (2009). Human mimicry. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 41, pp. 219–274). Elsevier.
  • Cheung, F., & Lucas, R. E. (2014). Assessing the validity of single-item life satisfaction measures: Results from three large samples. Quality of Life Research, 23(10), 2809–2818. https://doi.org/10.1007/s11136-014-0726-4
  • Clay, R. A. (2021). Mental health apps are gaining traction. Monitor on Psychology, 52(1). https://www.apa.org/monitor/2021/01/trends-mental-health-apps
  • Condon, W. S., & Ogston, W. D. (1967). A segmentation of behavior. Journal of Psychiatric Research, 5(3), 221–235. https://doi.org/10.1016/0022-3956(67)90004-0
  • Daly, M., Sutin, A. R., & Robinson, E. (2022). Longitudinal changes in mental health and the COVID-19 pandemic: Evidence from the UK household longitudinal study. Psychological Medicine, 52(13), 2549–2558. https://doi.org/10.1017/S0033291720004432
  • Davison, K. P., Pennebaker, J. W., & Dickerson, S. S. (2000). Who talks? The social psychology of illness support groups. American Psychologist, 55(2), 205–217. https://doi.org/10.1037/0003-066X.55.2.205
  • Delaherche, E., Chetouani, M., Mahdhaoui, A., Saint-Georges, C., Viaux, S., & Cohen, D. (2012). Interpersonal synchrony: A survey of evaluation methods across disciplines. IEEE Transactions on Affective Computing, 3(3), 349–365. https://doi.org/10.1109/T-AFFC.2012.12
  • Doré, B. P., & Morris, R. R. (2018). Linguistic synchrony predicts the immediate and lasting impact of text-based emotional support. Psychological Science, 29(10), 1716–1723. https://doi.org/10.1177/0956797618779971
  • Dragojevic, M., Gasiorek, J., & Giles, H. (2016). Accommodative strategies as core of the theory. In H. Giles (Ed.), Communication accommodation theory: Negotiating personal relationships and social identities across contexts (pp. 36–59). Cambridge University Press. https://doi.org/10.1017/CBO9781316226537.003
  • Dutta, M. J., & Feng, H. (2007). Health orientation and disease state as predictors of online health support group use. Health Communication, 22(2), 181–189. https://doi.org/10.1080/10410230701310323
  • Eichstaedt, J. C., Kern, M. L., Yaden, D. B., Schwartz, H. A., Giorgi, S., Park, G., Hagan, C. A., Tobolsky, V. A., Smith, L. K., Buffone, A., Iwry, J., Seligman, M. E. P., & Ungar, L. H. (2021). Closed- and open-vocabulary approaches to text analysis: A review, quantitative comparison, and recommendations. Psychological Methods, 26(4), 398–427. https://doi.org/10.1037/met0000349
  • Fortuna, K. L., Naslund, J. A., LaCroix, J. M., Biano, C. L., Brooks, J. M., Zisman-Ilani, Y., Muralidharan, A., & Deegan, P. (2020). Digital peer support mental health interventions for people with a lived experience of a serious mental illness: Systematic review. JMIR Mental Health, 7(4), e16460. https://doi.org/10.2196/16460
  • Gasiorek, J. (2016). Theoretical perspectives on interpersonal adjustments in language and communication. In H. Giles (Ed.), Communication accommodation theory: Negotiating personal relationships and social identities across contexts (pp. 13–35). Cambridge University Press.
  • Giles, H., Coupland, N., & Coupland, J. (1991). Accommodation theory: Communication, context, and consequence. In H. Giles, J. Coupland & N. Coupland (Eds.), Contexts of accommodation: Developments in applied sociolinguistics (pp. 1–68). Cambridge University Press; Editions de la Maison des Sciences de l’Homme. https://doi.org/10.1017/CBO9780511663673.001
  • Goldsmith, D. J. (2004). Communicating social support. Cambridge University Press.
  • Gonzales, A. L., Hancock, J. T., & Pennebaker, J. W. (2010). Language style matching as a predictor of social dynamics in small groups. Communication Research, 37(1), 3–19. https://doi.org/10.1177/0093650209351468
  • Guerrero, L. K., Jones, S. M., & Burgoon, J. K. (2000). Responses to nonverbal intimacy change in romantic dyads: Effects of behavioral valence and degree of behavioral change on nonverbal and verbal reactions. Communication Monographs, 67(4), 325–346. https://doi.org/10.1080/03637750009376515
  • Hatfield, E., Cacioppo, J. T., & Rapson, R. L. (1994). Emotional contagion. Cambridge University Press.
  • High, A. C., & Dillard, J. P. (2012). A review and meta-analysis of person-centered messages and social support outcomes. Communication Studies, 63(1), 99–118. https://doi.org/10.1080/10510974.2011.598208
  • Hipson, W. E., & Mohammad, S. M. (2021). Emotion dynamics in movie dialogues. PLOS ONE, 16(9), e0256153. https://doi.org/10.1371/journal.pone.0256153
  • Holt-Lunstad, J. (2017). The potential public health relevance of social isolation and loneliness: Prevalence, epidemiology, and risk factors. Public Policy & Aging Report, 27(4), 127–130. https://doi.org/10.1093/ppar/prx030
  • Holt-Lunstad, J., Smith, T. B., Baker, M., Harris, T., & Stephenson, D. (2015). Loneliness and social isolation as risk factors for mortality. Perspectives on Psychological Science, 10(2), 227–237. https://doi.org/10.1177/1745691614568352
  • Hove, M. J., & Risen, J. L. (2009). It’s all in the timing: Interpersonal synchrony increases affiliation. Social Cognition, 27(6), 949–960. https://doi.org/10.1521/soco.2009.27.6.949
  • Hu, M., & Liu, B. (2004). Mining and summarizing customer reviews [ Paper presentation]. Proceedings of the 2004 ACM SIGKDD International Conference on Knowledge Discovery and Data Mining – KDD ’04, Seattle, WA, USA. https://doi.org/10.1145/1014052.1014073
  • Ireland, M. E., Slatcher, R. B., Eastwick, P. W., Scissors, L. E., Finkel, E. J., & Pennebaker, J. W. (2011). Language style matching predicts relationship initiation and stability. Psychological Science, 22(1), 39–44. https://doi.org/10.1177/0956797610392928
  • Jockers, M. L. (2015). Syuzhet: Extract sentiment and plot arcs from text. https://github.com/mjockers/syuzhet>
  • Jones, S. M., Bodie, G. D., Youngvorst, L., Navarro, M., & Danielson, C. (2018). Mapping the terrain of person-centered supportive conversations. Communication Monographs, 85(4), 467–490. https://doi.org/10.1080/03637751.2018.1501503
  • Koole, S. L., & Tschacher, W. (2016). Synchrony in psychotherapy: A review and an integrative framework for the therapeutic alliance. Frontiers in Psychology, 7, 862. https://doi.org/10.3389/fpsyg.2016.00862
  • Koudenburg, N., Postmes, T., & Gordijn, E. H. (2017). Beyond content of conversation: The role of conversational form in the emergence and regulation of social structure. Personality and Social Psychology Review, 21(1), 50–71. https://doi.org/10.1177/1088868315626022
  • Lakey, B., & Orehek, E. (2011). Relational regulation theory: A new approach to explain the link between perceived social support and mental health. Psychological Review, 118(3), 482–495. https://doi.org/10.1037/a0023477
  • Lee, C. M., Cadigan, J. M., & Rhew, I. C. (2020). Increases in loneliness among young adults during the COVID-19 pandemic and association with increases in mental health problems. The Journal of Adolescent, 67(5), 714–717. https://doi.org/10.1016/j.jadohealth.2020.08.009
  • Lepore, S. J., Ragan, J. D., & Jones, S. (2000). Talking facilitates cognitive-emotional processes of adaptation to an acute stressor. Journal of Personality and Social Psychology, 78(30), 499–508. https://doi.org/10.1037/0022-3514.78.3.499
  • MacGeorge, E. L., & Zhou, Y. (2021). Social support, supportive communication, and health. In T. L. Thompson & N. G. Harrington (Eds.), The Routledge handbook of health communication (3rd ed. pp. 136–148). Routledge/Taylor & Francis Group.
  • Manson, J. H., Bryant, G. A., Gervais, M. M., & Kline, M. A. (2013). Convergence of speech rate in conversation predicts cooperation. Evolution and Human Behavior, 34(6), 419–426. https://doi.org/10.1016/j.evolhumbehav.2013.08.001
  • McCarthy, M. (2003). Talking back: “Small” interactional response tokens in everyday conversation. Research on Language and Social Interaction, 36(1), 33–63. https://doi.org/10.1207/S15327973RLSI3601_3
  • Mehl, M. R., & Pennebaker, J. W. (2003). The sounds of social life: A psychometric analysis of students’ daily social environments and natural conversations. Journal of Personality and Social Psychology, 84(4), 857–870. https://doi.org/10.1037/0022-3514.84.4.857
  • Mohammad, S. (2018). Obtaining reliable human ratings of valence, arousal, and dominance for 20,000 English words [ Paper presentation]. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, Melbourne, Australia (Vol. 1). Long Papers. https://doi.org/10.18653/v1/p18-1017
  • Mohammad, S., & Turney, P. (2010). Emotions evoked by common words and phrases: Using Mechanical Turk to create an emotion lexicon [ Paper presentation]. Proceedings of the NAACL HLT 2010 workshop on computational approaches to analysis and generation of emotion in text, Los Angeles, CA, USA (pp. 26–34).
  • Naslund, J., Aschbrenner, K., Marsch, L., & Bartels, S. (2016). The future of mental health care: Peer-to-peer support and social media. Epidemiology and Psychiatric Sciences, 25(2), 113–122. https://doi.org/10.1017/S2045796015001067
  • Nielsen, F. A. (2011). A new ANEW: Evaluation of a word list for sentiment analysis in microblogs [ Paper presentation]. Proceedings of the ESWC2011 Workshop on ‘Making Sense of Microposts’: Big things come in small packages 718 in CEUR Workshop Proceedings, Heraklion, Greece (pp. 93–98). http://arxiv.org/abs/1103.2903
  • Pennebaker, J. W., Boyd, R. L., Jordan, K., & Blackburn, K. (2015). The development and psychometric properties of LIWC2015. University of Texas at Austin.
  • Pew Research Center. (2021, April 7). Mobile fact sheet. Pew Research Center. https://www.pewresearch.org/internet/fact-sheet/mobile/
  • Pinheiro, J., Bates, D., DebRoy, S., Sarkar, D., & R Core Team. (2020). nlme: Linear and nonlinear mixed effects models. R Package Version 3.1-151. https://CRAN.R-project.org/package=nlme
  • Rains, S. A. (2016). Language style matching as a predictor of perceived social support in computer-mediated interaction among individuals coping with illness. Communication Research, 43(5), 694–712. https://doi.org/10.1177/0093650214565920
  • R Core Team. (2022). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/
  • Rinker, T. W. (2021). sentimentr: Calculate text polarity sentiment version 2.9.0. https://github.com/trinker/sentimentr
  • Roberts, N. A., Leonard, R. C., Butler, E. A., Levenson, R. W., & Kanter, J. W. (2013). Job stress and dyadic synchrony in police marriages: A preliminary investigation. Family Process, 52(2), 271–283. https://doi.org/10.1111/j.1545-5300.2012.01415.x
  • Rohrbaugh, M. J., Shoham, V., Butler, E. A., Hasler, B. P., & Berman, J. S. (2009). Affective synchrony in dual‐and single‐smoker couples: Further evidence of “symptom‐system fit”? Family Process, 48(1), 55–67. https://doi.org/10.1111/j.1545-5300.2009.01267.x
  • Sandstrom, G. M., & Dunn, E. W. (2014). Social interactions and well-being: The surprising power of weak ties. Personality and Social Psychology Bulletin, 40(7), 910–922. https://doi.org/10.1177/0146167214529799
  • Schöllgen, I., Morack, J., Infurna, F. J., Ram, N., & Gerstorf, D. (2016). Health sensitivity: Age differences in the within-person coupling of individuals’ physical health and well-being. Developmental Psychology, 52(11), 1944–1953. https://doi.org/10.1037/dev0000171
  • Silge, J., & Robinson, D. (2016). tidytext: Text mining and analysis using tidy data principles in R. The Journal of Open Source Software, 1(3), 37. https://doi.org/10.21105/joss.00037
  • Soliz, J., Giles, H., & Gasiorek, J. (2022). Communication accommodation theory: Converging toward an understanding of communication adaptation in interpersonal relationships. In D. O. Braithwaite & P. Schrodt (Eds.), Engaging theories in interpersonal communication (pp. 130–142). Routledge. https://doi.org/10.4324/9781003195511-12
  • Swartwood, R. M., Veach, P., Kuhne, J., Lee, H. K., & Ji, K. (2011). Surviving grief: An analysis of the exchange of hope in online communities. Omega: The Journal of Death & Dying, 63(2), 161–181. https://doi.org/10.2190/OM.63.2.d
  • Tausczik, Y. R., & Pennebaker, J. W. (2010). The psychological meaning of words: LIWC and computerized text analysis methods. Journal of Language and Social Psychology, 29(1), 24–54. https://doi.org/10.1177/0261927X09351676
  • Templeton, E. M., Chang, L. J., Reynolds, E. A., Cone LeBeaumont, M. D., & Wheatley, T. (2022). Fast response times signal social connection in conversation. Proceedings of the National Academy of Sciences, 119(4). https://doi.org/10.1073/pnas.2116915119
  • Tickle-Degnen, L., & Rosenthal, R. (1990). The nature of rapport and its nonverbal correlates. Psychological Inquiry, 1(4), 285–293. https://doi.org/10.1207/s15327965pli0104_1
  • van Uden Kraan, C. F., Drossaert, C. H. C., Taal, E., Lebrun, C. E. I., Drossaers-Bakker, K. W., Smit, W. M., Seydel, E. R., & van de Laar, M. A. F. J. (2008). Coping with somatic illnesses in online support groups: Do the feared disadvantages actually occur? Computers in Human Behavior, 24(2), 309–324. https://doi.org/10.1016/j.chb.2007.01.014
  • Warner, E. L., Ellington, L., Kirchhoff, A. C., & Cloyes, K. G. (2018). Acquisition of social support and linguistic characteristics of social media posts about young adult cancer. Journal of Adolescent and Young Adult Oncology, 7(2), 196–203. https://doi.org/10.1089/jayao.2017.0076
  • Weger, H., Castle Bell, G., Minei, E. M., & Robinson, M. C. (2014). The relative effectiveness of active listening in initial interactions. International Journal of Listening, 28(1), 13–31. https://doi.org/10.1080/10904018.2013.813234
  • Weisel, K. K., Fuhrmann, L. M., Berking, M., Baumeister, H., Cuijpers, P., & Ebert, D. D. (2019). Standalone smartphone apps for mental health—a systematic review and meta-analysis. NPJ Digital Medicine, 2(118). https://doi.org/10.1038/s41746-019-0188-8
  • Wickham, H. (2016). ggplot2: Elegant graphics for data analysis. Springer-Verlag New York.
  • Wind, T. R., Rijkeboer, M., Andersson, G., & Riper, H. (2020). The COVID-19 pandemic: The ‘black swan’ for mental health care and a turning point for e-health. Internet Interventions, 20, 100317. https://doi.org/10.1016/j.invent.2020.100317
  • Wright, K. (2000). Perceptions of on-line support providers: An examination of perceived homophily, source credibility, communication and social support within on-line support groups. Communication Quarterly, 48(1), 44–59. https://doi.org/10.1080/01463370009385579