Changes in social connection during COVID-19 social distancing: It’s not (household) size that matters, it’s who you’re with

To slow the transmission of COVID-19, countries around the world have implemented social distancing and stay-at-home policies—potentially leading people to rely more on household members for their sense of closeness and belonging. To understand the conditions under which people felt the most connected, we examined whether changes in overall feelings of social connection varied by household size and composition. In two pre-registered studies, undergraduates in Canada (NStudy 1 = 548) and adults primarily from the U.S. and U.K. (NStudy 2 = 336) reported their perceived social connection once before and once during the pandemic. In both studies, living with a partner robustly and uniquely buffered shifts in social connection during the first phases of the pandemic (βStudy 1 = .22, βStudy 2 = .16). In contrast, neither household size nor other aspects of household composition predicted changes in connection. We discuss implications for future social distancing policies that aim to balance physical health with psychological health.


Introduction
On March 11, 2020, the Centers for Disease Control and Prevention [1] declared the COVID-19 outbreak a pandemic. By early April, COVID-19 had already spread to nearly 1.5 million people worldwide [2]. In an effort to slow down its transmission, countries around the world implemented social/physical distancing guidelines [3], compelling individuals to stay at least 6 feet (2 meters) away from anyone outside their household [4]. Early in its implementation, the WHO [2] announced that it would be moving away from the term "social distancing" and begin using "physical distancing" to more accurately describe the practice. However, the original term stuck, especially in the U.S., U.K., Australia, Italy, France, Poland, Russia, India, South Korea, and Hong Kong, even though the very label of "social distancing" arguably undermines feelings of social connection. We use "social distancing" in this paper to reflect common usage. Such non-pharmaceutical public health interventions have been long proposed to reduce the spread of infectious disease. For example, mathematical modeling suggests that social distancing can reduce transmission of influenza by over 90% [5], and retrospective and be cognitively taxing [36]. In times of stress and crisis, these forms of online communication may in turn promote other negative outcomes, such as "Zoom fatigue" [37]. Thus, faceto-face interactions with household members are likely to be essential to increased feelings of social connectedness. In sum, social connection is crucial for both psychological and physical health, perhaps especially so during an unprecedented global pandemic that has claimed more lives than every war since the Korean War [38]. How can future policy guidelines balance protecting physical health through social distancing with protecting psychological health by maintaining feelings of connection? To understand the conditions under which people felt the most connected, we examined whether changes in overall feelings of social connection varied as a function of household size and composition.

Present research
In two pre-registered studies of undergraduates at a Canadian university (N Study 1 = 548) and adults primarily residing in the United States and United Kingdom (N Study 2 = 336), we followed individuals before and during the COVID-19 pandemic to examine changes in feelings of social connection based on (1) household size and (2) household composition. Using twotailed tests, we tested the following primary hypotheses. First, we expected that people in larger households to show relatively smaller declines (or bigger increases) in social connection as a result of the COVID-19 pandemic. Second, we hypothesized that household composition would predict changes in social connection as a result of the COVID-19 pandemic. In examining household composition, we focused on whether participants lived with a partner (or not), lived with a pet (or not), and were caregivers (or not). Feelings of social connection were assessed with three different measures-the Social Connectedness Scale (Study 1; [39]), the relatedness subscale of the Balanced Measure of Psychological Needs (BMPN; Study 2; [40]), and the UCLA Loneliness Scale (Study 2; [41]).

Study 1
Undergraduates at the University of British Columbia completed our measures as part of two separate surveys. We obtained ethics approval from the Behavioral Research Ethics Board at the University of British Columbia, and participants provided written consent to be part of our study. The first survey was completed before the COVID-19 pandemic (Time 1 ), and the second survey was completed during the COVID-19 pandemic (Time 2 ). We pre-registered our analysis plan and stopping rules on the OSF and they are available at [https://tinyurl.com/ yddwt28v]. A separate pre-registered study that used a portion of the data to answer a different research question can also be found on the OSF at [https://tinyurl.com/ybwz8ufb].

Time 1
Between January 6, 2020 and the end of March 2020, 3,504 participants completed demographic questions and a social connection measure alongside other items as part of an optional department-wide pre-screening. For consistency with Study 2, we only included participants who completed this questionnaire on or before February 12, 2020, resulting in a Time 1 sample of 2,903 students. After removing participants who were missing more than two items on the social connection measure (as pre-registered), we obtained a total sample of 2,708 eligible participants.

Time 2
We invited participants who had completed pre-screening at Time 1 to complete a second survey between April 1-8 th , 2020. The Time 2 survey consisted of the same measure of social connection as in the Time 1 survey, as well as measures assessing students' living arrangements, behaviors, and experiences during the COVID-19 pandemic. A total of 1,059 participants completed the Time 2 survey. As pre-registered, 8 participants were removed for responding 12 or more times in a row with the same answer on the social connectedness measure and 1 participant was removed for failing to answer more than 2 items on the social connectedness measure. Although not pre-registered, we also removed participants who did not supply an ID number to link responses (n = 125) or completed the survey twice (n = 22), For those who completed the survey twice, we only included their responses from the first survey.
Of the remaining 903 participants, 548 participants (M age = 20.78, SD age = 2.96; 77% women) completed both surveys and met our inclusion criteria. Participants in this final dataset did not significantly differ from the remaining eligible participants who completed the Time 1 survey in Time 1 social connection (p = .359) or household income (p = .154). Because we aimed to recruit as many participants as possible, we did not conduct an a priori power analysis; however, based on sensitivity analyses using GPower [42] and assuming two-tailed α = 0.05 and 80% power, we should have been able to detect a small effect size of ƒ 2 = .01 (R adj 2 = .02) in a 2-predictor regression model and ƒ 2 = .01 (R adj 2 = .001) in a 5-predictor regression model. R adj 2 is reported in the manuscript. The dataset for the final sample can be found on the OSF at [https://tinyurl.com/y7nvg5vf].

Measures
The measures for Study 1 can be found on OSF at [https://tinyurl.com/y7jfk4al].

Social connection
Social connection was assessed with the revised 20-item Social Connectedness Scale [39]. Participants indicated their level of agreement with items such as, "I feel close to people" and "I feel understood by the people I know" (1 = strongly disagree, 6 = strongly agree). We removed the item, "I feel comfortable in the presence of strangers" from both time points, because it may have had a different meaning in the midst of the pandemic. Participants completed the measure at Time 1 with reference to their general view of themselves (α = .94). At Time 2 however, due to the rapid changes to daily life that participants were experiencing, we asked them to think about the past week (α = .93).

Household size and composition
To assess household size, we asked participants "other than yourself, how many people are currently living in the same place you are now?" with answer choices ranging from "living alone" to "10+ people." For each person in their household, participants specified whether the person was a "spouse/partner/girlfriend/boyfriend" (subsequently referred to as partner), "child," "parent," "brother/sister," "other family member," "friend," "roommate/acquaintance," "livein help," or "other." Participants could only select one option per household member.

Living with pets
We asked whether participants were "currently living with any pets" (yes/no).

Being a caregiver
Participants were asked whether they were "currently the primary caregiver for anyone else (e.g., children or elderly family members)" (yes/no).

Social/physical distancing
Participants indicated whether they were "currently practicing social or physical distancing," and to indicate how many people aside from their household members got to within 6 feet or less of them on the previous day.

Hours spent video calling with family and friends
Participants were asked "yesterday, how many hours did you video call with family and friends" with answer choices ranging from "0" to "10+ hours."

Study 1 results
The code used to conduct the Study 1 analyses can be found on the OSF at [https://tinyurl. com/y7b8cnw3]. Correlations between all variables in Study 1 can be found in Table 1. Did household size buffer changes in social connection as a result of the COVID-19 pandemic?. As reported in Folk et al. [43], our sample exhibited a slight but significant decrease in feelings of social connectedness from Time 1 to Time 2 , and 98% of participants indicated they were social/physical distancing (see Table 2).
Pre-registered analyses. First, we examined whether household size (i.e., number of people in the household other than themselves) as a continuous measure (M = 2.54, range = 0 to 9 [with 77% living with 3 others or fewer], SD = 1.58) was associated with Time 2 social  Table 3, Model 1). We then examined the association between living alone and Time 2 social connectedness, controlling for Time 1 connectedness. In this model, living alone (n = 49) was not significantly associated with Time 2 connectedness b = -0.12, 95% CI = [-0.30, 0.07], p = .230 (see Table 3, Model 2).

Did household composition buffer changes in social connection as a result of the COVID-19 pandemic?.
Pre-registered analyses. While household size did not appear to play a role in changes in social connectedness from before to mid-pandemic, we investigated whether features of household composition were related to Time 2 connectedness. Controlling for Time 1 connectedness, living with a partner (n = 67) predicted significantly greater social connectedness at Time 2 , b = 0.22, 95% CI = [0.06, 0.38], p = .008 (see Table 3, Model 3). See  Table 3, Model 4). Although we also pre-registered a similar analysis investigating the effects of being a caregiver on social connection, we did not conduct it, as only 6 out of 548 participants reported being a caregiver. Exploratory analyses. To further investigate the relationship between household size and composition and Time 2 social connection, we entered the household variables (household size, living alone, living with a partner, and having a pet) into a single model predicting Time 2 social connectedness while controlling for Time 1 connectedness. Consistent with the results of our pre-registered analyses, in this full model, living with a partner was significantly associated with higher Time 2 social connectedness, b = 0.22, 95% CI = [0.06, 0.39], p = .008, whereas having a pet was significantly associated with lower Time 2 social connectedness, b = -0.15, 95% CI = [-0.26, -0.03], p = .014 (see Table 3, Model 5). No other effects were significant.
Was the relationship between household size and changes in social connection mediated by total hours video calling with family and friends or social distancing?. Pre-registered analyses. It is possible that we observed no relationship between household size and shifts in social connection because individuals in smaller households may be more likely to engage in video calling or may be less likely to socially distance from non-household members. However, correlations among these variables were nonsignificant (see Table 1), precluding mediation. The pre-registered mediation analyses are presented in S1 Table.

Study 2
Given our first study's reliance on college students, we sought to replicate its results with a sample of adults from around the globe (U.S., U.K., and 26 other countries), who were recruited to complete our survey at two timepoints: once prior (Time 1 ) and once during (Time 2 ) the COVID-19 pandemic. We obtained ethics approval from the Institutional Review Board at the University of California, Riverside, and participants provided written consent to join our study. Our pre-registered stopping rules and analysis plans for Study 2 are available at [https://tinyurl.com/y8s5ssm9] on the OSF website. A portion of the data was also included in another pre-registered study [https://tinyurl.com/yc8b2n44].

Time 1
On February 12, 2020, participants (N = 396; M age = 31.61, SD age = 11.88; 55% Male; 80% White; 46% single/never married; 32% U.S.; 27% U.K.) completed measures of social connection, loneliness, and demographics (along with other measures that were not part of our preregistered analysis plan). All participants were recruited from Prolific Academic TM , a recruitment platform demonstrated to provide quality online data [44].

Measures
The measures for Study 2 can be found on OSF at [https://tinyurl.com/yapg6tdt]. The same measures of (1) household size, (2) household composition (i.e., living with a partner), (3) living with pets, (4) being a caregiver, (5) social/physical distancing, and (6) hours spent video calling with family and friends were used as in Study 1.

Social connection
Social connection in this study was assessed with two measures: (1) the 6-item relatedness subscale of the BMPN [1,40] and (2) the 20-item UCLA Loneliness Scale [41]. The relatedness subscale asked participants to think about the past week and rate agreement with statements such as, "I felt close and connected with other people who are important to me" (1 = strongly disagree, 7 = strongly agree). Relatedness scores were highly reliable at both Time 1 (α = .76) and Time 2 (α = .77). The UCLA Loneliness Scale prompted participants to respond to statements based on how they feel in general (e.g., "People are around me but not with me"; 1 = never, 4 = often). Loneliness scores were highly reliable at both Time 1 (α = .88) and Time 2 (α = .88).

Hours spent working outside the home
Participants were additionally asked "how many hours per week do you work outside the home?" with answer choices ranging from "0" to "40+ hours."

Additional exclusion criteria
As pre-registered, to screen out inattentive participants, we planned to exclude those who provided the same answer 15 times in a row on the 20-item UCLA Loneliness Scale. We also preregistered to exclude those who were missing more than 1 item on the 6-item BMPN relatedness subscale and missing more than 2 items on the UCLA Loneliness Scale. However, we did not have any instances of inattentiveness or missing data.

Study 2 results
The R code used for the analyses in Study 2 can be found on OSF at [https://tinyurl.com/ y7nhpx7h]. Correlations among variables in Study 2 can be found in Table 4. Did household size buffer changes in social connection as a result of the COVID-19 pandemic?. As reported by Folk et al. (2020), our sample showed no changes in relatedness and small but significant improvements in loneliness from before to after the pandemic. Additionally, 93% of participants reported that they were social distancing (see Table 5).
Exploratory analyses. As in Study 1, we examined which aspects of household size and composition-when tested in a single model-best predicted Time 2 social connection after controlling for Time 1 social connection. None of the household size and composition variables were significantly associated with Time 2 loneliness, when controlling for Time 1 loneliness (see Table 7, Model 14). However, when we examined the same variables (household size, living alone, living with a partner, living with a child, living with a pet, and being a caregiver) in a single model predicting Time 2 relatedness, controlling for Time 1 relatedness, living with a partner was the only factor that buffered changes in social connection, b = .38, 95% CI. [0.09, 0.67], p = .012 (see Table 7, Model 13). This finding was consistent with Study 1.
Did working outside of the home moderate the effects of household size and composition on changes in social connection?. Pre-registered analyses. We expected that household size and household composition might matter less for social connection for individuals who worked outside the home. However, we did not find that hours working outside the home moderated the relationship between household size (continuous and living alone) or composition (living with a partner, living with children, living with a pet, being a caregiver) and changes in relatedness or loneliness (see S2 Table).
Was the relationship between household size and changes in social connection mediated by total hours video calling with family and friends or social distancing?. Pre-registered analyses. No significant correlations emerged between our outcome variable (relatedness, loneliness) and 1) our predictor variable (household size) and 2) our mediator variables (hours video calling, social distancing; see Table 4 for correlations). Thus, parallel to Study 1, the number of hours spent video calling with family and friends or social distancing did not mediate the relationship between household size (continuous and living alone) and changes in relatedness or loneliness (see S3 Table).

Discussion
Across two pre-registered studies that followed the same participants from before the COVID-19 pandemic into its early stages, we found that living with a partner was the strongest predictor of shifts in social connection across time. This finding replicated across two different samples-a sample of undergraduates at a Canadian university and a sample of adults from mostly the U.S. and the U.K. Both of our studies revealed robust positive regression coefficients indicating that people living with a partner were more likely to improve in social connection after social distancing guidelines were in place than those not living with a partner. This finding is consistent with past research demonstrating that being in a relationship is one of the strongest predictors of connection and well-being [11,45], in part because happier people are more likely to find partners [46,47]. Additionally, during times of worry and uncertainty, partners have been found to be more valuable for coping than other types of household members [26]. Moreover, recent research has shown that, on average, romantic relationships have not deteriorated over the course of the pandemic; indeed, people are relatively more willing to forgive their partners during COVID-19 [48]. In light of this evidence, it is not surprising that partners showed the strongest effect, especially during a pandemic. Contrary to our pre-registered hypotheses, changes in loneliness were not predicted by any other aspects of household composition. Furthermore, we found only nonsignificant trends for the impact of household size, including living alone, on social connection during COVID-19, perhaps because both our studies included small samples of those living in large households and households of one. It is important to keep in mind that the pandemic has forced people to spend unusually large amounts of time confined to home. Given that interpersonal interactions must be positive to contribute to one's overall sense of connectedness [10], those who live in larger households-relative to those who live alone or in smaller households-may have had more interactions that were negative (e.g., due to bickering or lack of privacy and alone time) and, as a result, failed to experience benefits in terms of social connection. Moreover, our studies measured experiences fairly early in the pandemic (April 2020); thus, as people continue to distance over long periods of time, their feelings of social connection may suffer. Going beyond household size and structure, future studies should examine the effects of relationship quality on social connection over time.
When examining how other features of household composition were associated with shifts in social connection during the pandemic, we obtained mixed findings regarding living with pets and null findings for all other household variables. However, because households are multifaceted, larger sample sizes will be needed to fully dissect the household composition findings, as well as to reveal interactions (such as with household size, gender, or country of residence). For example, studies with larger sample sizes may uncover differences in connection between those in households of four (with a partner and two children) versus households of five (with a partner and three children), and so on. Importantly, future investigators may wish to further unpack the role of household dynamics, as some households include unhealthy relationships that may be exacerbated by social distancing measures and others include housemates that minimally interact. As such, the quality and frequency of interaction among household members-perhaps with experience sampling or daily diary measures-is an important factor to explore in future work.

Implications and conclusions
Directed by social distancing interventions in the spring of 2020, millions of people were no longer commuting to work, attending school, or leaving their homes to spend time with friends and family. These extraordinary conditions likely led people to rely more on their household members to fulfill their needs for closeness, belonging, and connection [10]. The results from our two studies revealed that living with a partner-but not how many people or who else one lives with-appeared to confer unique benefits during these uncertain and unprecedented times. Indeed, demonstrating its robustness, this finding replicated across our two studies, despite weak and opposite correlations between household size and living with a partner (r = -.06 in Study 1 and .11 in Study 2). In light of these results, policy makers might consider developing guidelines for social/physical distancing that protect people's physical health while ensuring they retain a sense of closeness and connection by spending time in close proximity with partners, even outside their households. Some areas in the world, such as New Zealand, have implemented a strategy known as the "social bubble," which is the easing of social distancing to allow close contact with another household [49]. Such approaches might be especially helpful for individuals who have been unintentionally and disproportionally socially isolated by social distancing measures, such as those who are cut-off, separated from their partners, or generally struggling with staying at home. However, social bubbles pose a risk of increased infection rates [49]. Hence, just as safe sex education aims to reduce the rate of sexually transmitted diseases and unintended pregnancy, education on safe social distancing (or social bubbling) strategies might guide individuals across the globe how to connect with others safely while simultaneously curtailing COVID-19 rates. In sum, recommendations that reduce the risk of transmission while prioritizing social connection can ensure that people's physical and psychological health are optimally balanced.
Supporting information S1