A Dyadic Perspective of Felt Security: Does Partners’ Security Buffer the Effects of Actors’ Insecurity on Daily Commitment?

Interdependence and attachment models have identified felt security as a critical foundation for commitment by orientating individuals towards relationship-promotion rather than self-protection. However, partners’ security also signals the relative safety to commit to relationships. The current investigation adopted a dyadic perspective to examine whether partners’ security acts as a strong link by buffering the negative effects of actors’ insecurity on daily commitment. Across two daily diary studies (Study 1, N = 78 dyads and Study 2, N = 73 dyads), actors’ X partners’ daily felt security interactions revealed a strong-link pattern: lower actors’ felt security on a given day predicted lower daily commitment, but these reductions were mitigated when partners reported higher levels of felt security that day. Actors’ X partners’ trait insecurity (attachment anxiety) interaction also showed this strong-link pattern in Study 1 but not Study 2. The results suggest that partners’ felt security can help individuals experiencing insecurity overcome their self-protective impulses and feel safe enough to commit to their relationship on a daily basis.


Further Information on Study Sample
Dyadic studies are time and resource intensive, which has two important implications. First, sample sizes are necessarily constrained by funding and participant attrition, although the studies here were designed to overcome this limitation in prior research. Second, these studies are designed to examine multiple, independent processes. As outlined below, the aims, results and analyses presented in the current paper are independent of any reported in prior publications.

Consideration of Power
The sample sizes of Study 1 and 2 were determined based on prior research underpinning the multiple aims of broad projects examining actor and partner effects on daily dynamics. The target for each sample was 80 couples accounting for attrition due to non-compliance with the daily sampling procedure, which balanced funding with the aim to have adequate statistical power to detect small actor and partner effects based on prior studies and conventions at that time. However, the original aims of the study when funded did not specifically target actor X partner interactions. In addition, prior work on partner buffering effects indicating that partners might play a strong-link role have not examined these dyadic effects in daily interactions, and so we could not apply effect sizes from previous studies to consider a priori power analyses for these existing samples given the different measurement, context and analytic strategies across and within studies. Thus, we did not conduct a priori power analyses for the current investigation focused on actor X partner effects. In our Discussion section, we recognize the possibility that some effects may not have emerged due to limited power.

Prior Use of Samples
Prior papers using the data from Study 1 have focused on examining the degree to which individual factors entering the study, such as attachment insecurity and sexist attitudes, predict daily emotional and behavioral dynamics (Cross et al., 2019;Hammond & Overall, 2013;Overall et al., 2014Overall et al., , 2015Overall et al., , 2016, as well as the extent to which daily negative behaviors predict perceived partner responsiveness and relationship satisfaction Sasaki & Overall, 2020). Study 2 was specifically designed to replicate the design and procedure of Study 1. As in Study 1, prior papers using the data from Study 2 have focused on examining the degree to which individual factors entering the study, such as attachment insecurity and sexist attitudes, predict daily emotional and behavioral dynamics (Cross, et al., 2017;Girme, Overall, Simpson & Fletcher, 2015;Overall et al., 2015), as well as the extent to which daily negative behaviors predict perceived partner responsiveness and relationship satisfaction Sasaki & Overall, 2020). None of these prior studies have examined the degree to which actors' and partners' attachment insecurity and felt security are associated with commitment within daily life. Thus, all of the data and results presented in this paper are completely novel and independent of prior papers.

SPSS Syntax
We modelled the effects of within-person variations in (a) actors' felt security on day i, (b) partners' felt security on day i, and the (c) interaction between actors' and partners' felt security on day i predicting actors' commitment on day i, controlling for gender, the passage of time across the 21-day diary period, the corresponding between-person effects, and actors' commitment on day i-1 (Bolger & Laurenceau, 2013). The SPSS syntax is as follows: MIXED

Between-Person Effects
Table S1 presents the between-person effects in the same model that examined the withinperson effects of actors' felt security, partners' felt security and actor X partner felt security on actors' daily commitment. Unlike the within-person components that represent daily variations in actors' or partners' felt security, between-person components reflect individual differences in average levels of actors' or partners' felt security across days. In both studies, higher average levels of actors' felt security across days was associated with higher daily commitment. In Study 1, but not Study 2, higher average levels of partners' felt security was also associated with higher commitment. Moreover, in both studies, the between-person effects of actor X partner felt security were significant.
Accordingly, the higher commitment experienced by actors higher in felt security (right side of figure) was enhanced when partners were also higher in felt security Study 1: b = .36, t = 3.91, p < .001, r = .42; Study 2: b = .19, t = 2.29, p = .025, r = .27). Hence, the between-person effects also support that partners' high felt security can have beneficial effects, including buffering days of low insecurity (as shown by the central within-person effects focused on in the paper) and bolstering the positive between-person effects of actors' security on commitment across days. Note. The significant interaction effects presented in bold are presented in Figure S1. CI = confidence interval. Effect sizes (r) were computed using Rosenthal and Rosnow's (2007) formula: r = √(t 2 / t 2 + df). In these multilevel models, the Satterthwaite approximation is applied to provide specific degrees of freedom for each effect, which were used to calculate the effect sizes.

Figure S1. The between-person effects of Actor and Partner Felt Security on Daily
Commitment in Studies 1 and 2.
Note. Low and high levels of Actor and Partner felt security represent 1 SD below and above the mean. The simple effects of the slopes and contrasts are marked *** p < .001 and * p < .05.

SPSS Syntax
We modelled the effect of (a) actors' attachment anxiety, (b) partners' attachment anxiety, and the (c) interaction between actors' and partners' attachment anxiety predicting actors' daily commitment, controlling for gender, and the passage of time across the 21-day diary period (Kenny et al., 2006). The SPSS syntax is as follows:

CommD = Commitment
Day_c = day of diary, centered so that zero is midway through the 21-day diary period ATTANX_c = actor attachment anxiety, grand-mean centered p_ATTANX_c = partner attachment anxiety, grand-mean centered dyadid = unique dyad identifier obs = unique person identifier

Integrative Data Analyses
We conducted integrative data analyses by applying the same models examining RQ1 (as presented in Table 2) and RQ2 (as presented in Table 3) to the pooled data across both studies and included the main and interaction effects of study (-1 = Study 1, 1 = Study 2) to test whether the effects differed across studies. Table S2, study differences emerged in the main and interaction effects of actors' and partners' felt security on daily commitment. First, lower actors' felt security was associated with lower daily commitment, but this effect was stronger in Study 1 (b = .42, t = 22.49, p < .001, r = .32) than in Study 2 (b = .32, t = 16.55, p < .001, r = .24). Second, lower partners' felt security was associated with lower daily commitment in Study 2 (b = .10, t = 5.29, p < .001, r = .08),

RQ1. As shown in
but not in Study 1 (b = .01, t = .48, p = .632, r = .01). Third, the effects of actor X partner felt security differed across studies (as shown in bold in Table S2). Consistent with what was reported in the paper, a strong-link pattern was found in both studies, but the strong-link pattern was stronger in Study 2 than in Study 1 (as shown in Figure 1). Note. Analyses were conducted controlling for the corresponding between-person effects of felt security. CI = confidence interval. In these multilevel models, the Satterthwaite approximation is applied to provide specific degrees of freedom for each effect, which were used to calculate the effect sizes.
RQ2. As shown in Table S3, the effect of actor X partner attachment anxiety differed across studies (as shown in bold). Decomposing the interaction demonstrated a significant interaction revealing a strong-link pattern in Study 1 but not in Study 2 (similar to that shown in Figure 2).
Given that IDA revealed study differences in the effects of (1) actor X partner state felt security and (2) actor X partner trait insecurity, it is not appropriate to interpret the pooled effects from IDA. Instead, we present the effects for the two studies separately in the paper. Note. CI = confidence interval. Effect sizes (r) were computed using Rosenthal and Rosnow's (2007) formula: r = √(t 2 / t 2 + df). In these multilevel models, the Satterthwaite approximation is applied to provide specific degrees of freedom for each effect, which were used to calculate the effect sizes.

Gender Differences
To test whether gender moderated the effects of actors' and partners' felt security and attachment anxiety, we reran our analyses including all main and interaction effects with gender (-1 = women, 1 = men) separately for Studies 1 and 2.
Actor X Partner Felt Security. The effects were significant for both men and women, although there were some gender differences in the strength of the main effects. Gender differences emerged for the main effect of actors' within-person variations in felt security on daily commitment in Studies 1 (b = -.07, t = -3.34, p = .001, r = .07) and 2 (b = -.06, t = -3.36, p = .001, r = .08). Days of lower actors' felt security predicted lower commitment, but this effect was stronger for women (b = .47, t = 18.03, p < .001, r = .35 in Study 1 and b = .38, t = 17.05, p < .001, r = .37 in Study 2) than men (b = .34, t = 11.55, p < .001, r = .23 in Study 1 and b = .25, t = 8.22, p < .001, r = .19 in Study 2).

Differences across Relationship Length
We first tested whether relationship length was associated with daily felt security, attachment anxiety, and daily commitment. Only one significant effect emerged: longer relationship length was associated with lower attachment anxiety in Study 1 (b = -.01, t = -2.17, p = .033, r = .24). No other associations between relationship length, felt security and commitment emerged.
Next, to test whether relationship length moderated the effects of actors' and partners' felt security and attachment anxiety, we reran our analyses including all main and interaction effects with relationship length separately for Studies 1 and 2.

Anxiety
Simultaneously modelling the main and interaction effects of both (1) actors' and partners' daily felt security and (2) actors' and partners attachment anxiety on daily commitment produced identical results as when modelling felt security and attachment anxiety separately (see Table S4). .02 Note. Analyses were conducted controlling for the corresponding between-person effects of felt security. CI = confidence interval. Effect sizes (r) were computed using Rosenthal and Rosnow's (2007) formula: r = √(t 2 / t 2 + df). In these multilevel models, the Satterthwaite approximation is applied to provide specific degrees of freedom for each effect, which were used to calculate the effect sizes.

Testing Dyadic Effects of Attachment Avoidance on Commitment
Our paper focuses on attachment anxiety because it captures the tendency to appraise, experience and monitor felt security (Fraley & Shaver, 1998, 2000 that is key to our theoretical framework. In contrast, attachment avoidance reflects a motivational orientation that continually down-regulates closeness and dependence (rather than reflexively monitors felt-security to obtain closeness and dependence). While it is not part of the core aims of the current paper, the tendency for avoidant actors and partners to deal with insecurity by distancing from the other should also interfere with commitment (Mikulincer & Shaver, 2007;Pistole et al., 1995). Thus, we reran the analyses modelling the main and interaction effects of actors' and partners' attachment avoidance on daily commitment.
As shown in Table S5, across Studies 1 and 2, a main effect of actors' attachment avoidance on daily commitment emerged such that high actors' attachment avoidance was associated with lower daily commitment, but the actor X partner attachment avoidance was not significant. Simultaneously modelling the main and interaction effects of both (a) actors' and partners' attachment avoidance and (b) actors' and partners' attachment anxiety on daily commitment did not alter the effects of attachment anxiety (see Table S6). Note. CI = confidence interval. Effect sizes (r) were computed using Rosenthal and Rosnow's (2007) formula: r = √(t 2 / t 2 + df). In these multilevel models, the Satterthwaite approximation is applied to provide specific degrees of freedom for each effect, which were used to calculate the effect sizes. Note. CI = confidence interval. Effect sizes (r) were computed using Rosenthal and Rosnow's (2007) formula: r = √(t 2 / t 2 + df). In these multilevel models, the Satterthwaite approximation is applied to provide specific degrees of freedom for each effect, which were used to calculate the effect sizes.