Relationship difficulties and “technoference” during the COVID-19 pandemic

The COVID-19 pandemic has touched many aspects of people’s lives around the world, including their romantic relationships. While media outlets have reported that the pandemic is difficult for couples, empirical evidence is needed to test these claims and understand why this may be. In two highly powered studies (N = 3271) using repeated measure and longitudinal approaches, we found that people who experienced COVID-19 related challenges (i.e., lockdown, reduced face-to-face interactions, boredom, or worry) also reported greater self and partner phone use (Study 1) and time spent on social media (Study 2), and subsequently experienced more conflict and less satisfaction in their romantic relationship. The findings provide insight into the struggles people faced in their relationships during the pandemic and suggest that the increase in screen time – a rising phenomenon due to the migration of many parts of life online – may be a challenge for couples.


1) Analysis: within and between person analyses
We conducted additional exploratory analyses within-person centering the phone use variables to examine within-person variation across diary days and included their aggregate (i.e. each person's average across all diary days) in order to disentangle the withinand between-person differences (Bolger & Laurenceau, 2013;Zhang et al., 2009) between perceived phone use and relationship outcomes.
First, in separate models, we regressed relationship difficulties on own and partner phone use, and found a significant association at a within-person level (b = .10, SE = .03, 95% CI [.05, .16], p < .001) but not between-person (b = .06, SE = .06, 95% CI [-.05, .17], p = .283) for own phone use. The association was significant within-person (b = .09, SE = .03, 95% CI [.04, .14], p = .001) and between-person (b = .12, SE = .06, 95% CI [.01, .24], p = .039) for partner phone use. Next, in separate models, we regressed relationship satisfaction on relationship difficulties, controlling for either self or partner phone use. Results revealed a significant within-and between-person association between relationship difficulties and relationship satisfaction for both own phone use and perceived partner phone use. Indirect effects for within person models did not include zero, indicating that relationship difficulties mediated, in separate models, the negative relationship between own phone use, perceived partner phone use and relationship satisfaction. These analyses indicate that on days in which participants perceived greater phone use (their own or their partner's), they also experienced poorer relationship outcomes, consistent with the original between-person effects. Full results can be seen in Table 1. Given the wide variety of COVID-19 pandemic related policies, and their potential influence on people's experiences during the pandemic, using the same analytical design to test for mediation, we conducted additional analyses controlling for local social distancing policies, in both the concurrent and lagged models. We created a new "policy" variable related to the social distancing regulations in place in the previous two weeks (0 = no social distancing regulations, 1 = social distancing has been encouraged, 2= social distancing has been ordered, 3 = social distancing has been enforced by the police). We chose to control for social distancing policy as it is an index of how each location has handled the pandemic. Overall, in 33% of cases social distancing had been encouraged, in in 30.6% ordered, in 32.5% enforced, and in 3.9% of cases no social distancing policies had been reported. Controlling for this variable, the pattern of results remains the same, indicating that the personal experiences of the pandemic and their relationship with social media use and changes in relationship outcomes go above and beyond the actual policy measures. Second, social media was related to poorer relationship outcomes controlling for pandemic factors and local policy (and earlier relationship outcomes in the lagged models). The full tables of results are below. Table 2: Results of the mediation analysis in Study 2 on the effect of pandemic seriousness, pandemic threat, and time spent with friends, family, or colleagues, and boredom, on relational outcomes through earlier social media use, controlling for local policy. Social media use coefficients are path b in the mediation models, and were estimated controlling for path a of the predictor variable as well as local policy.
Predictor and effect

3) Analysis: within and between person analyses
First, we tested a model (with the same multilevel structure as in the original analyses) in which we examined the within person and between person effects crosssectionally, including, similarly to the above analyses for Study 1, the within-person centered predictor variable as well as its aggregate, to disentangle the within-and between-person differences (Bolger & Laurenceau, 2013;Zhang et al., 2009).
As expected, in separate models, we regressed social media use on each pandemic factor and found that social media use was significantly associated at both a between and within-person level with all pandemic factors: pandemic seriousness (between person: b =.14, SE = .02, 95% CI [. Next, in separate models, we regressed relationship conflict on social media use (within person and aggregate), controlling for each pandemic factor (within person and aggregate). Results revealed a significant association between social media use and conflict, but only at the between person level. That is, people who spent more time on social media compared to others (but not necessarily those who experienced greater fluctuations in their social media use over time) reported greater conflict with their partner. Indirect effects for between-person values were calculated and did not contain zero, indicating that social media use was a significant mediator between pandemic factors and relationship conflict in all models, at the between person level.
Finally, to test the full serial mediation, we tested whether social media use was related to relationship satisfaction through the mediating role of conflict. First, we regressed relationship conflict on social media use and found a significant between person association (b = .19, SE = .03, 95% CI [.13, .26], p < .001) but not within person (b = -.01, SE = .03, 95% CI [-.07, .06], p = .797). Second, we regressed relationship satisfaction on conflict controlling for social media use, and found a significant association. Again, indirect effect did not contain zero, indicating that social media use was related to lower relationship satisfaction through greater conflict with the partner at the between person level. These results indicate that people who experienced greater pandemic stressors, also engaged in greater social media use and reported worse relationship quality, but fluctuation in people's own social media use did not drive these effects. Full results can be seen in Table 4. Table 4: Results of the cross-sectional multilevel mediation analysis in Study 2 on the effect of pandemic seriousness, pandemic threat, and time spent with friends, family, or colleagues, and boredom on relational outcomes through social media use and relationship conflict. * indicates b pathways in the mediation models and were estimated controlling for a pathways of the predictor variables.

5) Power analyses
We conducted simulations for power analyses in R software (version 3.6.3) following the approaches recommended by Lane and Hennes (2018) and Kumle et al. (2020. We simulated 1,000 random data sets for Studies 1 and 2 capturing their structure. In each random sample the average correlations among predictors, percentage of missing data, and variables (predictors of technology use, social media use, relationship difficulties and relationship satisfaction) with fixed effect estimates, random intercept, and residual variance were set to be the same as in the original samples. We conducted three types of power analyses (post-hoc power analyses, analyses based on previous literature, and the SESOI method with analyses estimating the smallest effect size of interest detectable in a replication), as described below.
First, power analyses for actual results were conducted post-hoc. Although this approach is no longer recommended, we include it here as a comparison to the other approaches that have been used for multilevel models. As Table 1 shows (approach 1), Study 1's data showed 76% and 82% power, respectively, to detect the associations between lockdown and self phone and partner phone use, and 72% power between phone use and relationship difficulties. The link between relationship difficulties and satisfaction has >.80% power. In Study 2, data showed .77 and >.80% power to detect the associations between social media use and relationship difficulties, and between relationship difficulties and satisfaction, respectively. Regarding social media use predictors, while pandemic serious, threat and boredom showed >. 80%, the connection between times spent with friends, family, and colleagues, showed between 42-100% power.
Second, we tested how much power the data sets for Studies 1 and 2 have to detect reasonable effects based on previous literature (approach 2). Unstandardized effects sizes from prior research appear in Table 1. Specifically, we used McDaniel and Drouin's (2019) effect to test the association between technology use and relationships difficulties, and McDaniel et al. (2020) for the relationship of difficulties and satisfaction, given that these works seem to represent the general effect sizes reported in the relevant literature. Given the recency of pandemic and its unique characteristics, it is more difficult to determine what a reasonable expected effect size would be for the lockdown  phone use pathway in Study 1. After careful consideration, we chose papers that examined internet and social media use before and after public health crises in China (Lyu, 2012) and before and during lockdown in Italy (Cellini et al., 2020), given that the context of crisis and the study designs in these papers largely reflected the conditions in our Study 1, with phone use assessed before and during lockdown. Results revealed that Study 1 had between 63% and 87.2% power to predict phone use. For Study 2, we used Stockdale and Coyne's (2020) effects of the motives for technology use (connection, information seeking and boredom) for our estimations, as they better mirrored the variables and design used in our study as predictors of social media use. Results revealed that pandemic seriousness, threat, boredom, and time spent with family, friends, and colleagues have >.80% power to predict social media use. Finally, both Studies 1 and 2, have >.80% power to detect the associations between technology use and relationship difficulties, and between relationship difficulties and satisfaction based on prior research effects. Thus, based on estimates from previous literature, our studies appear to generally have adequate power to detect the effects of interest.
Third, given that the prior two approaches to estimating power are subject to potential biases (e.g. even adopting effect sizes from published data involves the risk of performing the analyses on inflated effect sizes, Kumle et al., 2020), we also estimated power by progressively decreasing the estimates of the focal associations in our Studies until power was 80% (approach 3). This approach enabled us to identify the smallest effect sizes that could be reliably detected in a replication study with the same sample size and parameters as those used in this project. The results revealed that a replication of Study 1 would provide 80% power to detect an unstandardized association of .38 between predictor of technology use and phone use (1.23% more than the association observed in the present study), and between predictor of technology use and partner phone use (3% less than the association observed in the present study), 81% power to detect an unstandardized association of .11 between phone use and relationship difficulties (1.22% more than the association observed in the present study), and 85% power to detect an unstandardized association of between relationship difficulties and satisfaction .10 (79.2% less than the association observed in the present study). For Study 2, a replication study would have a 85% power to detect an unstandardized association of .045 between predictor of technology use and social media use (71.9-43.75% less than the association observed in the present study; excluding the predictor time spent with family which need an 2.25 more), 81% power to detect an unstandardized association of .10 between social media use and relationship difficulties (1.1 % than the association observed in the present study), and 86% power to detect an unstandardized association of -.04 between relationship difficulties and satisfaction (85.2% less than the association observed in the present study). These results indicate that, overall, a replication using the same sample size and method would have adequate power (>.80%) to detect effect sizes similar (if not even smaller) than the ones found in our studies.