To Text or Talk in Person? Social Anxiety, Media Affordances, and Preferences for Texting Over Face-To-Face Communication in Dating Relationships

ABSTRACT Collocated dating couples have the choice between interacting in person or via communication media, inviting theorizing about when and why they gravitate toward mediated communication. Drawing on O’Sullivan’s (2000) model, we investigated young adults’ preference for texting over face-to-face (FtF) communication in situations varying in threat (mundane talk vs. difficult conversations vs. breakups), as a function of individuals’ social anxiety and perceptions of media affordances. Results show that higher social anxiety predicted preferences for texting, and this relationship got stronger as relational episodes were more threatening. The perceived importance of editability mediated the relationship between social anxiety and preferences for texting under threatening circumstances. The results shed light on the factors that drive media preferences in relational contexts.

Gone are the days when face-to-face (FtF) was people's primary means of social interaction.Communication media, such as texting, e-mail, and videoconferencing, have become widely used in all aspects of individuals' personal lives (McClain et al., 2021).This high-choice media environment makes it possible to inquire about people's preferences: When FtF and mediated communication are both available, what factors drive individuals toward using the media and why?Understanding media preferences is important not only because people tend to act upon their preferences, but also because they reveal people's understanding of how the media can be leveraged to achieve their communicative goals -a key issue for media scholars (Hartmann, 2009).What media attributes do people consider important in various contexts and why?What are their mental calculations when orienting toward the media?
A prominent theory of media preferences is O'Sullivan's model of strategic channel use in close relationships (O'Sullivan, 2000).This theory argues that the key driver of preferences for mediated over FtF communication in this context is communicators' goal to manage interpersonal threat, or situations that challenge their ability to convey desirable impressions or to manage personal relationships effectively.Mediated communication is argued to be preferable under these taxing circumstances because it allows individuals more control over message composition, thus increasing their likelihood of attaining their interaction goals.
In the present article, we apply O'Sullivan's model to understand media preferences during romantic episodes involving threat -breakups and difficult conversations -amongst individuals in geographically collocated dating relationships.We focus on texting because it is the most frequently used medium in this relationship context, with dating couples spending far more time texting than using any other media (e.g., Toma & Choi, 2016).More importantly, texting is a highly controllable medium, thus providing an ideal setting for testing O'Sullivan's propositions.As described in detail later, we aim to make theoretical contributions to O'Sullivan's model on two primary fronts.First, individual differences that affect media preferences under threatening circumstances need further specification in O'Sullivan's model.In the current study, we consider the role played by individuals' social anxiety in their preferences for texting over FtF, thus conceptualizing media preferences as dually shaped by situational factors (i.e., the presence of interpersonal threat) and individual characteristics relevant to this situation (i.e., social anxiety).Second, O'Sullivan's model proposes that the ambiguity inherent in many media facilitates information control, which explains individuals' preferences for mediated interaction.However, the role of the media in supplying this control has not been directly examined.Here, we draw on recent theorizing on media affordances (e.g., Evans et al., 2017, Rice et al., 2017) to specify the media-based mechanism responsible for individuals' preferences for texting vs. FtF communication in this context, thus aligning O'Sullivan's model with the latest developments in the field of computer-mediated communication (CMC).In sum, we ask: how do individuals' social anxiety and perceptions of media affordances shape their preferences for texting vs. FtF when negotiating romantic episodes involving threat?

O'Sullivan's Impression Management Model of Strategic Channel Use in Close Relationships
O' Sullivan's (2000) model conceptualizes communication media as tools that people can access in their strategic pursuit of relational goals, focusing on the media's usefulness during situations involving interpersonal threat.The model's core proposition is that individuals in close relationships turn to mediated channels in their efforts to manage interpersonal threat, such as when they are confronted with an accusation or must confess to a transgression.In these situations, relational management becomes harder, and the media can provide unique advantages.Specifically, mediated communication is inherently more ambiguous than FtF communication, due in part to its reduction in nonverbal cues, thus offering individuals more leeway to control information to their advantage.For example, it is easier to mask unattractive aspects of self and highlight desirable ones in typed messages than in spontaneous FtF discussions (O'Sullivan, 2000).
Empirical support for these assertions emerged in several studies.In O'Sullivan's foundational study (2000), participants were asked to recall positive and negative interactions with their romantic partner, and then report their preference for using mediated communication over FtF to deliver their messages.Compared to episodes in which participants praised their interaction partner or highlighted positive aspects of self, episodes involving threat (either confessing to or accusing the partner of a transgression) led to a shift in preferences from FtF to mediated communication.In a similar study using hypothetical scenarios, individuals who imagined making a confession about a relational transgression to a casual friend showed greater preferences for media with limited cues (e.g., texting; Ruppel, 2018), providing support for the notion that having one's own face threatened intensifies preferences for mediated vs. FtF communication.
Research has also found a shift from FtF to mediated channels in scenarios involving high interpersonal risk (and hence the potential for interpersonal threat to arise), such as asking a manager for a pay raise or asking an acquaintance out on a date (Joinson, 2004).In a direct replication of O'Sullivan's study, Feaster (2010) asked participants to recall a recent threatening encounter and report their preferences for several media as well as the extent to which they felt confident in their ability to control information in each medium.Participants' self-reported abilities for information control in each medium predicted their preferences for using that respective medium.For instance, those who felt more confident controlling the information delivered via texting were more likely to prefer to text under threatening episodes.This finding supports O'Sullivan's contention that it is individuals' perception of the controllability afforded by mediated channels that drives media preferences.

Threatening Relational Episodes
We now apply O'Sullivan's model to romantic episodes involving interpersonal threat: breakups and difficult conversations.Breakups occur when one or both romantic partners register irreconcilable concerns with the relationship and decide to end it (Field et al., 2009).In college-aged populations, romantic breakups are ranked as some of the most distressing experiences, akin to grief, for both initiators and receivers (Field et al., 2009).While the distress experienced by the rejected party is easy to imagine, breakup initiators also experience substantial negative emotions, such as guilt and shame at the prospect of hurting their former partner; lower self-esteem due to perceiving the relationship as a failure; and anxiety due to anticipating stigma and lack of support from their social circles (Duck, 2007).Breakup conversations are thus perceived as threatening and difficult to accomplish, a feeling exacerbated by scriptlessness, or breakup initiators' lack of knowledge of what do say or do to alleviate their partner's distress (Baumeister & Dhavale, 2001).
The second threatening episode is difficult conversations.In dating relationships among young adults, there can be much uncertainty about the future of the relationship, fear of rejection, jealousy, and worries about infidelity (Lantagne & Furman, 2017).The process of addressing these issues has been labeled difficult conversations (Donovan, 2015), and has been shown to be often accompanied by intense negative emotions, such as frustration, anger, and resentment (Pietromonaco et al., 2004).However, unlike breakups, individuals enter difficult conversations with the goal of preserving the relationship and often succeed in doing so, with these conversations enhancing intimacy within the couple (Pietromonaco et al., 2004).Thus, the key difference between breakups and difficult conversations lies in the conversation initiator's intent: either to terminate the relationship (i.e., breakups) or simply to manage the relationship considering the fraught issue (i.e., difficult conversations).
Additionally, we investigate media preferences during a non-threatening relational episode: mundane talk.Mundane talk refers to the daily chatter of everyday life, where romantic partners talk about their daily activities, engage in simple coordination of tasks, or casually discuss issues of personal interest (Overall et al., 2010).While O'Sullivan's model is not designed to explicate media preferences during non-threatening encounters, mundane talk provides a useful comparison group against which individuals' media preferences for threatening episodes can be contrasted.In summary, breakups and difficult conversations can be conceptualized as threatening relational episodes, with the former more threatening than the latter, while mundane talk is a nonthreatening episode.According to O'Sullivan's model, we predict that preferences for texting vs. FtF communication should be proportional with the level of threat experienced in each relational episode: H1: Participants' preference for texting over FtF will be highest for breakups, followed by difficult conversations, and least of all mundane talk.

Social Anxiety
While O'Sullivan 's model (2000) theorizes media preferences as shaped primarily by interpersonal threats, he himself acknowledged that the role of individual differences relevant to how individuals respond to threatening situations requires further specification within the model.Here, we consider social anxiety as such an individual-level predictor of preferences for mediated communication.Social anxiety is defined as the extent to which individuals worry about and fear being evaluated negatively in social situations (Leary & Kowalski, 1997), and thus should be directly implicated in how they handle interpersonal threat.Socially anxious individuals tend to doubt their ability to make positive impressions upon others, believe themselves to have low social skills, and expect others to respond negatively to them (Leary & Kowalski, 1997).As a result of these perceptions, socially anxious individuals tend to experience excessive nervousness and exhibit difficulties in managing nonverbal behaviors, such as by allowing long silences, averting eye gaze, or requiring longer response time than less socially anxious individuals.In the area of personal relationships, higher social anxiety has been associated with poor functioning.Social anxiety inhibits individuals from maintaining regular social contact, or enacting relationship maintenance behaviors; in conflict situations, high anxiety individuals exhibit reduced assertiveness and increased conflict avoidance, precisely because of their fear of negative evaluation (e.g., Davila & Beck, 2002).
Given these difficulties with managing social situations and especially threatening ones, we expect individuals with higher social anxiety to exhibit a consistent preference for mediated, as opposed to FtF interaction, for all the reasons delineated by O'Sullivan's model: less need to manage nonverbal cues, more control over message construction, more opportunities to process messages.The literature on the effects of psychosocial well-being on preferences for mediated interaction supports this contention.Research in this paradigm argues that psychosocial problems increase preferences for online interaction for similar reasons as those articulated by O'Sullivan's model, namely that individuals with these afflictions perceive mediated communication as more comfortable and less threatening than FtF communication (e.g., Reid & Reid, 2007, Shalom et al., 2015).For example, Caplan's model of problematic Internet use argues that individuals with higher social anxiety prefer online communication because it alleviates their social fears, a claim that received support in several survey studies (e.g., Caplan, 2007).Similarly, the social compensation hypothesis argues that individuals with psychosocial vulnerabilities compensate for these vulnerabilities by using mediated communication (Valkenburg & Peter, 2007, Toma, 2022).This claim has received strong support in the context of social anxiety (e.g., Reid & Reid, 2007, Shalom et al., 2015).Hence, we hypothesize: H2: Across the three types of relational episodes, participants with higher social anxiety will display a greater preference for texting over FtF.
In the original formulation of O'Sullivan's model, the presence of interpersonal threat was the sole predictor of individuals' preference for mediated communication.In this expanded version, that considers social anxiety as an additional predictor, interpersonal threat should operate by altering how socially anxious individuals orient toward the media.Recall that, by definition, social anxiety affects how individuals respond to threat (Davila & Beck, 2002).Since individuals with social anxiety have a high discomfort with threatening social situations and mediated communication offers some reprieve from this discomfort, they should display a greater preference for mediated communication the more threat they experience.Thus, interpersonal threat should moderate the relationship between individuals' social anxiety and their preference for mediated communication.In the present study, breakups should inspire the greatest preference for texting, because they are the most threatening scenario, followed by difficult conversations, and least of all mundane talk: H3: The positive association between participants' social anxiety and their preference for texting over FtF will be strongest for breakups, followed by difficult conversations, and least of all mundane talk.

Media Affordances
As previously reviewed, O'Sullivan (2000) explains individuals' preferences for mediated over FtF communication in threatening situations by arguing that the absence of nonverbal cues in mediated environments creates ambiguity; in turn, "the opportunities for ambiguity that [mediated environments] provide can be harnessed in individuals' efforts to regulate self-relevant information" (p.408).Simply put, the model argues that the ambiguity present in mediated environments heightens users' ability to control information to their advantage.Indeed, one study finds that individuals' preferences for mediated communication were explained by their perceptions of information control within each medium (Feaster, 2010).But what characteristics of mediated communication supply this control?Recent CMC scholarship has argued that an affordance-centric approach is best suited for understanding the role played by media technology in social interaction (Evans et al., 2017, Rice et al., 2017).Below, we briefly review the affordance perspective and discuss how O'Sullivan's model can be brought into alignment with it.
The concept of affordances originates from the field of ecological psychology, where it was defined as the possibilities for action offered by objects in the environment (Gibson, 1979).It was then co-opted by design scholars (e.g., Norman, 1988), who viewed everyday objects as containing features that afford useful deployment.For instance, a chair "affords" sitting or a door "affords" opening.CMC scholars have conceptualized affordances as media features that users perceive as relevant to their goals and needs, focusing on how the materiality of technologies (i.e., the objective features that technologies possess) are subjectively noticed and interpreted by users (Treem & Leonardi, 2013).Recently, affordances have gained prominence as a particularly useful lens for theory-building in the field of CMC (Evans et al., 2017, Rice et al., 2017, Ronzhyn et al., 2022), for two main reasons.They are granular, meaning that they identify the specific elements of CMC environments that are implicated in social processes, thus allowing scholars to theorize the mechanisms behind CMC effects.They are also platformagnostic, meaning that they exist across platforms and are much more enduring than the platforms themselves.Thus, as media platforms change, become defunct, or newly emerge, CMC theories based on affordances can maintain their relevance.
To align O'Sullivan's model with this valuable, affordance-centric perspective, we translate O'Sullivan's arguments about the importance of information control into affordances.Following Evans et al. (2017)'s conceptualization, affordances cannot be a social or psychological outcome, but rather perceptions of technological features that lead to that particular outcome.In our case, information control is the outcome, and we argue that the following affordances generate this outcome.
First, bandwidth refers to users' ability to access nonverbal cues, such as body language, proxemics, and haptics, when communicating with their partners (Daft & Lengel, 1986, Fox & McEwan, 2017).FtF communication provides the highest bandwidth, offering access to the full range of nonverbal cues in addition to verbal cues.By comparison, all mediated communication reduces users' ability to access nonverbal cues (and therefore bandwidth), with some media completely eliminating them (e.g., texting, e-mail), while others only offering vocalics (e.g., phone) or vocalics and body language (e.g., videoconferencing).A long line of scholarship has argued that text-based mediathose completely eliminating nonverbal cues -are more controllable than "richer" media, where users are able to access nonverbal cues to some degree (Walther, 1996(Walther, , 2007)), for several reasons.Nonverbal cues are often produced unconsciously and are notoriously difficult to regulate, even when one is motivated to do so, such as when trying to avoid being caught lying (DePaulo et al., 2003).By contrast, verbal information (i.e., what one has to say) can be prepared in advance, polished, rehearsed, and revised -in a word, controlled.Additionally, thoughtfully controlling nonverbal cues requires cognitive resources, thus reducing the resources available for message construction (Walther, 2007).Simply put, it is mentally harder to come up with the right things to say when simultaneously trying to control one's body language, than when there is no body language to worry about.Relatedly, cognitive load is increased when individuals monitor their partners' nonverbal cues, which also takes away cognitive resources from message composition.The reduction of nonverbal cues has also been conceptualized as a "controllable affordance" (Toma, 2016), because it offers users greater latitude to construct desirable messages.
Second, editability refers to users' ability to revise messages before sharing them.It is virtually impossible to retract a statement made during FtF interaction.However, people can easily delete or polish messages before sending them via mediated channels.Editability is literally what allows individuals, in the words of O'Sullivan (2000), "to shade, shape, and vary the amounts of selfrelevant information as they balance the potential costs and rewards of revealing or concealing the many bits of information that constitute aspects of themselves" (p.424).Research shows that individuals who were more highly motivated to make positive impressions on online communication partners engaged in more editing behaviors, such as deletions, insertions, and replacements of words, and reported greater mindfulness while composing their messages (Walther, 2007).
In sum, scholarship to date shows that reduced nonverbal cues (i.e., low bandwidth) and editability are sources of information control, enabling online communicators to compose thoughtful messages, that are aligned with their communicative goals.In the context of O'Sullivan's model, low bandwidth and high editability can then be construed as explanatory variables for why individuals gravitate toward mediated communication when they are faced with interpersonal threat.
At the operational level, these affordances have been measured in two ways by recent scholarship: either as individuals' perceptions of the presence of affordances in communication media (e.g., Fox & McEwan, 2017, Mao & DeAndrea, 2019), or as individuals' perceptions of the importance of affordances in various communication contexts (e.g., Abeele et al., 2017, Choi & Toma, 2021).Both approaches focus on users' perceptions of technological features, rather than the objective existence of these features, which is aligned with current conceptualization of affordances (e.g., Evans et al., 2017).Here, we adopt the latter approach, because we seek to understand to what extent individuals believe that media affordances are relevant to their psychological needs, especially needs that stem from social anxiety, not simply the extent to which individuals notice the presence of affordances.

Mechanism Behind Preferences for Texting: Full Conceptual Model
So far, we have drawn on O'Sullivan's model to hypothesize that relational episodes involving interpersonal threat will produce greater preferences for mediated communication.We have then argued that O'Sullivan's model can be expanded by including (1) social anxiety as an individual-level predictor of preferences for mediated communication that affects how people respond to threatening circumstances and (2) media affordances (i.e., low bandwidth, high editability) as mechanistic variables that explain the attractive qualities of the media under threatening circumstances.Now, we assemble these variables (i.e., the presence of interpersonal threat, social anxiety, media affordances) into a full conceptual model that describes our proposed expansion to O'Sullivan's model.
As detailed in H2, we expect that individuals' social anxiety will have a direct effect on their preference for texting, such that higher anxiety will produce greater preferences for texting.As reviewed, socially anxious individuals fear negative evaluation and have trouble regulating their own nonverbal behaviors.Hence, mediated communication, where information can be more carefully controlled, should be especially appealing to them.By extension, the technological affordances that supply information control (i.e., low bandwidth, editability) should be the reason why socially anxious individuals prefer mediated communication to a greater extent than their less anxious counterparts.Thus, the direct effect of social anxiety on preferences for texting should be mediated by controllable affordances: H4: Individuals with higher social anxiety will display greater preferences for texting because they place more importance on controllable affordances (i.e., low bandwidth, high editability).
As detailed in H3, we expect that the relationship between individuals' social anxiety and their preference for texting will be moderated by relational episode (i.e., the presence of threat), such that higher anxiety individuals' preference for texting will be stronger the more threat they encounter.As previously argued, the reason texting is preferable for handling threatening situations is because it contains controllable affordances that make it easier to compose thoughtful messages and distance oneself from the discomfort of face-to-face confrontations.Individuals higher in social anxiety should then place more importance on having access to controllable affordances the more threat they encounter, since these affordances help them navigate situations that they find increasingly harder to navigate.In other words, relational episode (i.e., the presence of threat) should moderate the relationship between individuals' social anxiety and the importance they place on controllable affordances: H5: The positive association between social anxiety and the importance of controllable affordances (i.e., low bandwidth, high editability) will get stronger for increasingly more threatening relational episodes.
Finally, we expect that the relationship between controllable affordances and individuals' preference for texting should also be moderated by interpersonal threat.That is, the comfort and reassurance provided by controllable affordances should give rise to preferences for texting in general, because people should prefer the media that contain affordances they perceive as important.However, under conditions of threat, controllable affordances should be the single most salient factor that shapes people's preference for texting.Thus, the relationship between the perceived importance of controllable affordances and preferences for texting should be stronger when individuals experience interpersonal threat than when they don't.Consistent with this reasoning, one recent study (Chen & Lu, 2023) finds that when individuals perceive certain affordances as important, they prefer to use media that contain those affordances, and this relationship becomes stronger when they experience stressful situations, where those affordances provide a strategic advantage.Hence: H6: The importance individuals place on controllable affordances (i.e., low bandwidth, high editability) will be positively associated with their preferences for texting, and this relationship will get stronger for increasingly more threatening relational episodes.
In sum, in our expansion of O'Sullivan's model, we theorize that preferences for mediated communication can be understood as stemming from individuals' personal characteristics that influence how they respond to interpersonal threat (i.e., social anxiety), the amount of interpersonal threat they encounter, and the perceived importance of affordances that provide comfort and reassurance during these threatening encounters.Taken together, we propose a full conceptual model where individuals' social anxiety increases their preferences for texting because it provides access to controllable affordances (i.e., low bandwidth, high editability), and the links between (a) individuals' social anxiety and the importance they ascribe to controllable affordances; and (b) the importance they ascribe to these affordances and their preferences for texting get stronger the more individuals are faced with threat (see Figure 1).

Participants and Recruitment
Participants (N = 257, M age = 20.48,SD age = 1.70, 79% women, 91% heterosexuals, 51% singles, 77% White, 20% Asian or Asian American, 2% Black or African American) were recruited from the undergraduate research pool at a large Midwestern university and were compensated with course extra-credit.
All students above 18 years old were eligible for participation, regardless of whether they were currently in a romantic relationship.Two participants who did not complete the full survey were excluded from the final analysis.The sample size was determined through an a priori power analysis using G*Power.For a repeated measures design with one within-factor (see details in the following section), a sample of 195-252 participants was indicated to identify a small effect size of .10 with a statistical power of .80 ~ .90.We ran a few extra participants to account for any data loss during data collection.

Procedure and Design
Data were collected through an online survey administered via Qualtrics.Following O'Sullivan's methodology (see also Feaster, 2010, Ruppel, 2018), participants were presented with a series of hypothetical scenarios occurring in romantic relationships and were asked to report their interest in using texting and FtF to communicate with a romantic partner in each situation.All participants were asked to imagine interactions with a collocated romantic partner they had been dating exclusively for 6 months but were not cohabiting with.Participants completed a battery of questions about their hypothetical media use and the perceived importance of media affordances in these interactions.Finally, participants completed an exit questionnaire that included measures of individual differences and demographics.
A repeated-measure experimental design was used.The within-subjects factor, type of relational episode, had three levels: mundane talk, difficult conversations, and breakup.In the mundane talk condition, participants imagined four commonplace daily conversations with a romantic partner, such as making dinner arrangements or discussing the weather.In the difficult conversations condition, participants imagined four threatening conversations, such as revealing to the partner that they had received romantic attention from an attractive person.Finally, in the breakup condition, participants imagined only one conversation in which they decided to break up with the imagined partner due to not being happy in the relationship.The scenarios were adopted from previous research on romantic relationship maintenance (Goldsmith, & Baxter, 1996) and are presented verbatim in Table 1.In the mundane talk and difficult conversations conditions, the order in which the scenarios were presented was randomized.Additionally, the order of these three conditions was randomized.Plausibility and manipulation checks were administered after each of the three types of relational episode.

Measures
Media use.After reading each scenario, participants were asked to indicate, on a scale from 1 (strongly dislike) to 7 (strongly like), to what degree they "would like to use" FtF communication and then texting for communicating with their romantic partner.Recall that the mundane talk and difficult conversation conditions contained four scenarios each.Thus, reliability was calculated for participants' media use ratings for the four scenarios in each condition (for interest in FtF, α = .74~.80; for interest in texting, α = .63~.77).To validate our measurement, confirmatory factor analyses (CFA) were conducted with media use ratings for FtF and texting separately under the three conditions, revealing acceptable model fit (for FtF, χ 2 (25) = 42.122,p = .017,CFI = .95,RMSEA = .07,SRMR = .05;for texting, χ 2 (25) = 38.772,p = .039,CFI = .90,RMSEA = .06,SRMR = .07).Thus, participants' ratings of FtF and texting for mundane talk and difficult conversations were averaged into indices, respectively.The breakup condition only included one scenario; therefore, no reliabilities were computed.In line with O'Sullivan's (2000) procedure, new scores indicating participants' preference for texting over FtF (hence forth, preference for texting) in each condition were calculated by subtracting the texting score from the FtF score.Again, a CFA using this preference score showed good model fit, where the hypothetical scenarios loaded together under each condition as expected, χ 2 (25) = 39.23,p = .035,CFI = .95,RMSEA = .07,SRMR = .07(see Supplementary Information for a detailed report).
Perceived importance of media affordances.On a scale from 1 (completely disagree) to 7 (completely agree), participants indicated the extent to which two media affordances (bandwidth and editability) were important to them when communicating about the three relational episodes.Two subscales were adapted from Fox and McEwan's (2017) well-validated scale of media affordances for social interaction.The bandwidth subscale included 4 items (e.g., "It is important that this channel allows me to receive cues about how the other person is feeling.,"α = .87~.90, for mundane talk, M = 5.93, SD = .82;for difficult conversations, M = 6.31,SD = .74;for breakups, M = 6.41,SD = .88).The higher the composite bandwidth score, the more important participants thought it was to have access to nonverbal cues during the relational episode.For editability, the subscale also included 4 items (e.g., "It is important that this channel allows me to carefully craft my message before sending it," α = .88~.94,for mundane talk, M = 4.80, SD = 1.30; for difficult conversations, M = 5.03, SD = 1.40; for breakups, M = 4.98, SD = 1.64).
The higher the editability score, the more important participants thought it was to be able to edit their messages.
Social anxiety.Social anxiety was measured with the short form of the Social Interaction Anxiety Scale (SIAS: Peters et al., 2012).Participants reported to what extent they agreed with six statements representing their regular social interaction behaviors on a 5-point Likert scale (e.g., "I have difficulty talking with other people;" 1= not at all characteristic of me, 5 = extremely characteristic of me, α = .88,M = 1.91,SD = 0.80).
Covariates.We accounted for age, gender, relationship status, and sexual orientation, because these variables might affect people's behaviors and perceptions in dating relationships.Participants reported their age in years, their gender as female (coded as 1) or male (coded as 0) (no participants selected other gender options), their relationship status as single (coded as 0) or in a committed relationship (coded as 1), and their sexual orientation as heterosexual or straight (coded as 1) or lesbian, gay, or bisexual (coded as 0).

Analytic Approach
Since we used a repeated-measures design, with participants making media use ratings for each type of relational episode, our data were non-independent.Thus, we analyzed it through a series of linear mixed-effects models (LMEM) using the lme4 R package (Bates et al., 2015).LMEM is highly recommended for analyzing non-independent data, because it accounts for within-person variances and minimizes both type-I and type-II errors, unlike other inferential analyses (Brauer & Curtin, 2018).In each LMEM model, participants were specified as a random factor.The LMEM analysis provided fixed effects following the standard interpretation of a generalized linear model (see Table 2).To test the full conceptual model (i.e., H4-6), we used the mediation package in R (Tingley et al., 2014).All analyses controlled for age, gender, relationship status, and sexuality.

Descriptive Statistics
To understand participants' interest in FtF and texting, we begin by reporting descriptive statistics for these variables separately.For mundane talk, participants exhibited a high interest in both texting (M = 5.84, SD = 0.82) and FtF (M = 6.06,SD = 0.84).As expected, participants reported higher interest in FtF than texting for mundane talk, t(256) = 4.12, p < .001,d = .26.For difficult conversations, participants' interest in FtF remained high (M = 6.50, SD = 0.72), but their interest in using texting (M = 3.83, SD = 1.50) was significantly lower, t(256) = 23.58,p < .001,d = 1.47.The same pattern emerged for breakups: Participants exhibited high interest in using FtF (M = 6.58,SD = 1.05), but a much lower interest in using texting (M = 2.54, SD = 1.93), t(256) = 24.87,p < .001,d = 1.55.Overall, our sample preferred FtF over texting across all relational episodes considered.The difference between the two interaction venues was relatively small for mundane talk, but quite large for difficult conversations and breakups (see Figure 2).Correlations among all main variables are reported in Supplementary Information.

Test of H1-H3
H1-H3 investigated how individuals' preferences for texting are dually shaped by the level of interpersonal threat they experience (i.e., situational factor) and their social anxiety (i.e., individual-level factor).In an LMEM, preference for texting was entered as the outcome variable, and social anxiety, relational episode, and their interaction terms as predictors.Since the difference between the levels of the relational episode variable is unknown, this categorical variable was treated as ordinal (Field et al., 2017).H1 hypothesized that preferences for texting will grow stronger from mundane talk to difficult conversations to breakups.As shown in Table 2, a statistically significant, negative association between type of relational episodes and preference for texting emerged, b = -1.908,p < .001.Given the ordinal nature of the relational episode variable (where mundane talk = -1, difficult conversations = 0, breakups = 1), this negative relationship indicates that preferences for texting decreased as threats heightened across relational episodes.According to the joint hypothesis test, coefficients significantly differed between difficult conversations and breakups, χ 2 = 679.42,p < .001.That is, participants' preference for texting was highest for mundane talk, followed by difficult conversations, and least of all breakups, rejecting H1.
H2 proposed that participants' social anxiety would be positively related to their preferences for texting over FtF.As predicted, a statistically significant positive association emerged, b = 0.483, p < .001,indicating that the more socially anxious participants were, the stronger preferences for texting they displayed, after accounting for control variables.This supported H2.We then predicted that the positive association between social anxiety and preference for texting would be highest for breakups, followed by difficult conversations, and then mundane talk (H3).The interaction between social anxiety and type of relational episode was statistically significant, b = 0.378, p < .001,indicating that the positive association between social anxiety and preference for texting became stronger as threats increased.Follow-up analyses confirmed significant differences between these coefficients, χ 2 = 714.43,p < .001.In other words, the higher participants' social anxiety, the more they preferred to use texting for breakups, followed by difficult conversations, and least of all mundane talk.Thus, H3 was supported.

Test of H4-6
H4-6 were concerned with the mechanism through which individuals' social anxiety was associated with their preference for texting.To begin with, H4 addressed an indirect relationship between social anxiety and preference for texting via media affordances, such that the more socially anxious individuals were, the more individuals would value low bandwidth and high editability, which in turn would predict their preference for texting (see Figure 3).For this mediation hypothesis, several LMEMs were needed.First, two separate LMEMs were built, with the perceived importance of bandwidth and the perceived importance of editability as the outcome variable, respectively.In both LMEMs, social anxiety, relational episode, and their interaction terms were entered as predictors.As shown in Table 2, social anxiety was positively associated with editability, b = .280,p = .003,but not associated with bandwidth, b = -.126,p = .100.***p < .001,**p < .01,*p < .05.The type of relational episode was person-mean centered for easy interpretation (i.e., mundane talk = -1, difficult conversations = 0, breakup = 1).For gender, female = 1, male = 0 (no participants selected other gender options); for relationship status, single = 0, in a committed relationship = 1; for sexual orientation, heterosexual = 1, lesbian, gay, or bisexual = 0.
This means that, across relational episodes, the more socially anxious participants were, the more they placed importance on editability.
To test the second part of H4 (i.e., the association between affordances and preference for texting), another LMEM was used with preference for texting as the outcome variable and the perceived importance of bandwidth and editability as predictors.As expected, bandwidth was negatively associated with preference for texting, b = -.439,p < .001,meaning that the less participants valued bandwidth, the more they preferred texting.By the same token, editability was positively associated with preference for texting, b = .218,p < .001,meaning that the more participants valued editability, the more they preferred texting.A formal test of mediation found a significant indirect effect via editability, with an estimated effect of .061(95% CI = [.021,.110],p < .001),meaning that the more socially anxious individuals were, the more they valued editability, which in turn predicted their preferences for texting.No significant indirect effect emerged for bandwidth (95% CI = [-.010,.090],p = .120).Taken together, these findings partially supported H4.
H5-H6 then addressed the moderating role of relational episodes.H5 hypothesized that relational episode would moderate the positive association between social anxiety and the importance of controllable affordances, such that increasingly more threatening relational episodes would strengthen higher-anxiety-individuals' perception that controllable affordances are important.Here, we examined the two LMEMs with editability and bandwidth as outcome variables respectively, where the interaction between social anxiety and relational episodes was considered as the predictor.As shown in Table 2, there were no significant moderating effects of relational episodes on the association between social anxiety and affordances (for the model with Figure 3. Indirect effect of media affordances in the relationship between social anxiety and preference for texting.***p < .001,**p < .01,*p < .05.Indirect effects were tested via the mediation R package (Tingley et al., 2014) based on 1,000 stimulations using quasi-Bayesian confidence intervals.ACME refers to average causal mediation effects (i.e., the effect size of indirect effects).The type of relational episode was person-mean centered for easy interpretation (i.e., mundane talk = -1, difficult conversations = 0, breakup = 1).
bandwidth as the outcome, b = .001,p = .997,and for the model with editability as the outcome, b = .028,p = .631).This suggests that individuals with higher social anxiety consistently valued the ability to edit messages regardless of how threatening the situation was.Therefore, H5 was rejected.
H6 hypothesized that the positive associations between perceived importance of controllable affordances and preferences for texting would grow stronger for increasingly more threatening relational episodes.We investigated the LMEM with preferences for texting as the outcome and the interaction terms between affordances and relational episode as predictors.Significant interaction effects emerged for bandwidth*relational episode, b = -.273,p = .001,and for editability*relational episode, b = .149,p = .002.Since relational episode was treated as an ordinal variable, these coefficients indicate that the negative relationship between bandwidth and preferences for texting became stronger as threats increased across relational episodes (see the Supplementary Information for plots that deconstruct these interaction effects.).Similarly, the positive relationship between editability and preferences for texting became stronger as threats increased.Thus, H6 was supported.Considering the results for H4-H6 together, socially anxious individuals attached higher importance to editability, which in turn predicted their preferences for texting.This association between editability and preference for texting got stronger for increasingly more threatening relational episodes.

Additional Analyses
Consistent with O'Sullivan's (2000) operationalization, the preference for texting variable used in all analyses reported earlier was computed as a difference score between interest in using texting and interest in using FtF.However, the procedure of using a difference score has been criticized because it leads to loss of variance in the original variables (e.g., Bonate, 2000).To alleviate this concern, we re-ran all the analyses above with interest in using texting and FtF entered separately in LMEMs.The findings were consistent with the results above and reported in Supplementary Information.

Discussion
Drawing on O'Sullivan 's model (2000), the current study investigated preferences for texting relative to FtF communication in a dating context involving threatening (difficult conversations, breakups) and non-threatening (mundane talk) relational episodes.Results show that, in an absolute sense, participants preferred FtF communication over texting to communicate with dating partners across relational episodes.Nonetheless, these preferences shifted in certain situations and for certain individuals.Preferences for texting decreased as relational episodes got more threatening -they were lowest for breakups, followed by difficult conversations, and least of all mundane talk.But this pattern was reversed when considering individuals' social anxiety.The higher their social anxiety, the more they preferred texting, and this relationship got stronger as relational episodes were more threatening.Next, we investigated the role played by participants' perceptions of media affordances in their preferences for texting across relational episodes.As expected, the more participants valued high editability and low bandwidth, the more they preferred texting.The more socially anxious individuals were, the more they valued high editability.In fact, the value they placed on high editability (but not bandwidth) mediated the relationship between social anxiety and preferences for texting, and this relationship got stronger for more threatening relational episodes.Taken together, some of these results provide support for O'Sullivan's model, while some do not, suggesting important boundaries for the utility of the model.O'Sullivan's (2000) model provides a cohesive, systematic way of understanding individuals' preferences for mediated vs. FtF communication within interpersonal relationships.These preferences are important to investigate because they reveal people's understanding of how the media functions and how it can be coopted to facilitate interaction success.The model's core prediction is that individuals are savvy about their communication options and gravitate toward those that allow them to put their best foot forward.Mediated communication, especially the textbased kind, provides greater control over message production than FtF, and thus is theorized to be preferred when circumstances require careful message construction, such as when dealing with interpersonal threats.

O'Sullivan's Model Revisited
Our main goal was to test and expand this model in several ways.Our first contribution was to test the model's predictions in the context of pivotal, defining moments in the lifespan of romantic relationships (i.e., breakups and difficult conversations), unlike previous research which has focused on isolated incidents involving threats, such as making accusations or confessing to specific transgressions (e.g., Joinson, 2004, Ruppel, 2018).A direct application of O'Sullivan's model would predict that breakups, the most threatening episode, would engender higher preferences for texting than difficult conversations, and both episodes would result in higher preferences for texting than mundane talk, which doesn't involve any threat.However, the data did not support this prediction.In fact, the opposite pattern emerged.In the context of pivotal moments in a couple's life, preferences were squarely in favor of FtF communication, with respondents indicating that they valued having access to nonverbal cues.
This finding suggests potential updates to O'Sullivan's model.First, social meaning ascribed to texting and other mediated channels might have changed since O'Sullivan's model was published two decades ago.Now social norms may dictate that in pivotal relational moments, couples should speak FtF.For instance, research shows that young people consider breaking up over text as unacceptable, likely because texting can be perceived as an impersonal venue for holding such a consequential conversation (Lenhart et al., 2015).Future research on O'Sullivan's model should consider incorporating social norms as an additional predictor of media preferences.Second, relational investment theories (Canary & Stafford, 2001) suggest that the brevity involved in texting may be perceived as indicating low investment in the relationship, which may sabotage the successful resolution of difficult conversations.Thus, pivotal romantic episodes might require the use of rich media as a signal of relational investment, unlike the isolated conversations investigated by prior research.Third, research on romantic conflict suggests that texting may be perceived as presenting disadvantages for relational management.For instance, individuals report strongly disliking having to wait for a partner's response via texting, with longer response latencies escalating feelings of anger (Scissors & Gergle, 2016).It is possible that these considerations (e.g., the symbolic meaning of media, the importance of signaling relational investment, and the distaste for delayed responses) outweigh individuals' interest in careful interaction management during pivotal relational episodes, and lead them to prefer FtF communication.Our unexpected finding regarding preferences for texting in the context of pivotal relational episodes thus suggests intriguing avenues for expanding and refining O'Sullivan's model, that we hope will be pursued by future research.
Our second contribution to O'Sullivan's model was to incorporate an individual-level factor as a predictor of preferences for mediated vs. FtF communication -social anxiety.Media choices are likely to be determined by a constellation of situational, individual, and media factors (see Chen & Lu, 2023, Choi & Toma, 2021), whereas O'Sullivan's model is narrowly focused on a single situational factor (i.e., the presence of interpersonal threat).Expanding the model to include individual characteristics relevant to how threat is experienced has been a key theoretical goal outlined by O'Sullivan himself (2000).We focused on social anxiety, because by definition it affects how individuals respond to interpersonal threat, such that highly anxious individuals experience even stronger discomfort with threat and are more likely to seek refuge from it (Leary & Kowalski, 1997).The inclusion of social anxiety proved fruitful.Despite the unexpected main effect of relational episode discussed earlier, O'Sullivan's propositions on information control received clear support when individuals' social anxiety was taken into consideration.The more anxious individuals were, the more they preferred texting for threatening episodes (breakups, followed by difficult conversations) compared to non-threatening episodes (mundane talk).In fact, the more anxious individuals were, the more they gravitated toward texting in general.When accounting for individuals' social anxiety, O'Sullivan's predictions were supported verbatim, suggesting that the desire to control messages may override the other considerations outlined earlier (e.g., social norms) for highanxiety individuals.Future research should directly investigate these possibilities.
Our final contribution to O'Sullivan's model lies in conceptualizing and testing the affordance-based mechanism responsible for preferences for texting.While O'Sullivan's model proposed that it is the controllability of information that drives individuals toward the media and away from FtF under conditions of threat, it did not account for the specific media characteristics that supply this controllability.We incorporated recent theorizing on media affordances to do so and found that higher social anxiety was linked with more importance placed on editability, which, in turn, was linked with higher preferences for texting; importantly, these associations got stronger the more individuals experienced threat, suggesting that situational demands, such as dealing with threat, render editability more central in individuals' media preferences.This affordance-based mechanism indicates that users are aware of the properties of the media that benefit them in relational encounters, and thus lends credence to O'Sullivan's claims that individuals are strategic in their utilization of communication media to meet psychological needs.One intriguing finding is that only editability, but not reduced nonverbal cues (i.e., low bandwidth), explained why socially anxious individuals gravitated toward texting during threatening relational episodes.Editability seems to be key to formulating desirable messages, and merits greater attention in future studies.It is worth noting, though, that it is possible that our operationalization of bandwidth did not adequately capture the theoretical dynamics outlined by O'Sullivan's model.Our measurement, which we adapted from Fox and McEwan (2017), captures the importance of having access to cues, which could be conceptually different than the importance of obscuring cues.We invite future research to develop and deploy a scale focused on the latter phenomenon, because it more accurately captures O'Sullivan's propositions.
Our focus on media affordances responds to recent calls in the scholarly community to carefully attend to media affordances, which provide a theoretically precise explanation of how people use and are affected by the media (see Evans et al., 2017).Yet limited research has tested this affordancebased mechanism, an issue that has spurred much attention from scholars lately: "the concept of affordances has been increasingly applied to the study of information and communication technologies [. ..].However, almost no research operationalizes affordances, limiting comparisons and programmatic research" (Rice et al., 2017, p. 106).Here, we built on the incipient body of research that operationalizes affordances (e.g., Abeele et al., 2017, Choi & Toma, 2021), focusing on individuals' perceptions of the importance of affordances during communication episodes, as opposed to their perceptions of the mere presence of affordances.Our approach proved empirically fruitfully, with individuals' perceptions of the importance of controllable affordances, especially editability, yielding statistically significant associations with both their media preferences and their psychological needs to manage social anxiety and to handle threat.Future research might benefit from operationalizing affordances in a similar way.

Preference for Texting or for Face-To-Face Interaction?
In contrast to O' Sullivan's (2000) predictions, our findings show that individuals exhibited higher interest in using FtF than texting across all three relational episodes.Additionally, when not accounting for social anxiety, participants' interest in using texting decreased as relational episodes became more threatening, while their interest in using FtF remained high and stable across situations.Thus, in a strict sense, participants preferred communicating with dating partners FtF by and large, and moved away from texting in emotionally-laden situations (difficult conversations, breakups).As previously discussed, in the specific context of negotiating pivotal episodes in romantic relationships, several factors may make texting undesirable (for instance, participants' distaste for long response latencies, and the fact that texting signals low relationship investment).
A close inspection of the data reported in O'Sullivan's foundational study (2000) and Feaster's (2010) replication reveals a similar pattern where participants' "default" preference was for FtF communication, not the media.This pattern of findings is consistent with a small body of research that suggests FtF is perceived as the "gold standard" when it comes to connecting with relational partners (see Dienlin et al., 2017).Interview studies find that, while both texting and FtF are perceived as convenient, FtF is perceived as more likely to lead to intimate connections, feeling understood, and socializing effectively (e.g., Robinson & Stubberud, 2012).Research on actual media use similarly shows that, when sharing significant emotional events, FtF is much more frequently used than any other media, even when sharing negative news (Choi & Toma, 2014).Likewise, when engaging in conflict with close relational partners, individuals are far more likely to do so FtF, although the use of mediated communication has been increasing (Scissors & Gergle, 2016).These findings do not necessarily mean that mediated communication is not functional, effective, or widely used in social interactions.Rather, they suggest that scholars might benefit from considering individuals' strong preferences for FtF communication, at least in interactions with close relationship partners.

Limitations and Additional Directions for Future Research
Several limitations need to be acknowledged.We used hypothetical scenarios where participants were asked to imagine interactions with a romantic partner, a procedure that is typical in the literature on O'Sullivan 's model (e.g., Feaster, 2010, Ruppel, 2018).One limitation of hypothetical scenarios is reduced ecological validity, although we carefully pretested them and ensured they had high plausibility.Future studies could benefit from implementing experience sampling or daily diary procedures, where participants track their own media preferences in vivo, as they engage in relational encounters.Participants in our study always reported their interest in using face-to-face first, and then their interest in using texting.Future studies should randomize the order in which participants rate the media, in order to avoid primacy effects.The hypothetical scenarios we used referenced generic difficult conversations and only included one breakup scenario.Future research should investigate more nuanced characteristics of these threatening episodes, such as their level of severity or the locus of responsibility for any transgressions discussed.
An equally valuable research direction is examining the extent to which the media preferences delineated by O'Sullivan's model translate into media use behaviors.Do individuals end up using their preferred media in threatening interactions with relational partners?Research suggests that people often act on their media preferences (Hartmann, 2009), but media use behaviors can also be influenced by the availability of the media (Choi & Toma, 2014), normative use (Treviño et al., 2000), or the partner's preferences (Yuan et al., 2016).
Our study used a sample of undergraduate students where women were disproportionally represented.Future research should replicate our findings with more balanced gender samples, and with individuals across age groups and relational stages (e.g., marital relationships) to increase generalizability.Future research should also seek to include participants who experience more severe social anxiety.The present findings are encouraging, in that strong association between social anxiety and preference for texting emerged even through our sample reported somewhat low mean scores on this variable.The role of social anxiety might be even stronger for participants who report higher anxiety.This study only focused on texting because it is by far the most widely used medium in college students' dating relationships.Future research should include additional media, such as phone calling or videoconferencing, not only because considering these media increases ecological validity, but also because, as previously discussed, these media may contain novel affordances of theoretical interest.

Figure 2 .
Figure 2. Descriptive statistics for participants' interest in FtF and texting across the three relational episodes.

Table 1 .
Hypothetical scenarios for each type of relational episode.You would like to share with your romantic partner your thoughts about today's weather.2.You would like to touch base with your romantic partner about tonight's dinner plan.3.You would like to share your regular school day with your romantic partner before going to bed. 4.You would like to let your romantic partner know that you met a mutual friend today.Difficult conversations 1.You really like your romantic partner, and you would like to move in together with him/her.But you are not sure how he/she feels about this, and you would like to discuss the possibility of moving in together.2.You attended a party by yourself, and an attractive person paid lots of attention to you.You would like to discuss what happened at the party with your romantic partner.3.You found out that your romantic partner did not tell your mutual friends that he/she was dating you.You are thus upset with him/her and would like to discuss it.4.You and your romantic partner have been in a relationship for a month, but now you both are going back to your respective home states for summer break.You would like to know where the relationship is going.Breakup episode 1.You and your romantic partner have been in a relationship for 6 months.You are not happy with the relationship, so you decide to break up and would like to let your romantic partner know.

Table 2 .
Linear mixed-effects models predicting media affordances and preference for texting.