Social relationship provisions and loneliness in school: Child- and classroom-level effects

Abstract

Building on social needs theory (Weiss, 1974), this study introduces the construct of classroom provision richness and examines the association between the exchange of social provisions among children in classrooms and children’s feelings of loneliness in school. We examined the receipt of provisions from reciprocally nominated friends versus unilateral (one-sided) and non-friend classmates and examined associations between social provisions and loneliness at the child and classroom levels. Participants were 998 third- through fifth-grade children (468 girls, 530 boys; 88.5% White) in 38 classrooms who indicated which classmates they played with, helped, validated, and provided opportunities for self-disclosure. In addition to the social provisions nomination measure, children responded to (a) a measure of loneliness that avoided content overlapping with social provisions, (b) a rating-scale sociometric measure of peer acceptance, and (c) a measure that asked them to indicate which classmates engaged in prosocial, aggressive, or withdrawn-type behaviors. Multilevel analyses indicated that social provisions received from reciprocal friends and from unilateral–received friends were associated with children’s feelings of loneliness in school. Furthermore, a measure of the provision richness of classrooms moderated the association between child-level provisions received and feelings of loneliness, such that children who received fewer provisions were less lonely in classrooms that were more provision-rich. Classroom provision richness was also associated with the general level of prosocial behavior and peer acceptance in the classroom. Together, findings suggest that efforts to foster the exchange of social provisions in classrooms could reduce loneliness and facilitate a more caring classroom environment.

Introduction

Loneliness is the emotional experience of feeling alone, cutoff, or distanced from others, arising from a perceived discrepancy between one’s desired and achieved levels of social connection (Parkhurst & Hopmeyer, 1999; Perlman & Peplau, 1981). Feelings of loneliness can be reliably assessed as early as kindergarten entry (e.g., Cassidy & Asher, 1992; Coplan et al., 2007), and approximately 10% of children report frequently experiencing loneliness in elementary school (Asher et al., 1984; Galanaki, 2004). Loneliness has implications for well-being throughout the lifespan (e.g., Hawkley & Cacioppo, 2010), and loneliness in middle childhood (approximately ages 8–11 years) is linked with adjustment concurrently and over time. For example, loneliness during elementary and middle school has been associated with concurrent academic self-concept and achievement (Levitt et al., 1994) and has been linked longitudinally with symptoms of anxiety and depression in adolescence (Fontaine et al., 2009). Furthermore, children whose loneliness increased across middle childhood were reported to experience higher levels of social withdrawal and peer victimization than those who remained low or decreased in loneliness during this time (Jobe-Shields et al., 2011). Additionally, high levels of loneliness in middle childhood have been linked with sleep problems and poorer general health in adolescence, even when high levels of loneliness were not maintained into adolescence, suggesting that loneliness in middle childhood may have lasting health effects (Harris et al., 2013).

Various indicators of how children are doing in their peer relations have been consistently related to loneliness during middle childhood and adolescence (see Schwartz-Mette et al., 2020, for a meta-analysis), likely reflecting the increasing amount of time children spend with peers across this period (Ellis et al., 1981; Lam et al., 2014). Research with children and adolescents indicates that being accepted by classmates, having friends in class, and having high-quality friendships each uniquely contribute to lower levels of loneliness (see Asher & Paquette, 2003, Rotenberg & Hymel, 1999, Schwartz-Mette et al., 2020, and Weeks & Asher, 2012, for reviews). Peer relationships also play a significant role in various indicators of children’s adjustment to school more broadly (Ladd et al., 1996; Ryan & Shin, 2018; Wentzel, 2017).

Much of the research on friendship quality, in particular, is grounded in a social needs theoretical perspective on loneliness. Weiss (1974) and Shaver and Buhrmester (1983) proposed that the social provisions people receive from their close relationships are a key protective factor against feelings of loneliness, meeting social needs for intimacy, attachment, and social integration. By social provisions, researchers refer to the interpersonal benefits derived from social relationships, such as companionship, help, validation, and opportunities for self-disclosure.

In the present study, we assessed for the first time the exchange of social provisions among all peers in the classroom during middle childhood and examined how the receipt of provisions overall and from different sources (e.g., mutual friends, unilateral friends, non-friends) are related to children’s feelings of loneliness in school. In addition, we examined provision exchange at the classroom level—a classroom characteristic we refer to as classroom provision richness—and how this feature of the classroom environment may be associated with loneliness, as well as considering classroom provision richness as a potential moderator of the association between child-level provisions received and children’s feelings of loneliness in school.

Social Provisions and Loneliness

Consistent with a social needs perspective, children and adolescents whose closest friendship provides more social provisions are less lonely than children whose closest friendship is less provision-rich (Bukowski et al., 1993; Ladd et al., 1996; Lodder et al., 2017; Nangle et al., 2003; Parker & Asher, 1993). Given the importance of this single closest friendship for loneliness, it is understandable that much of the focus in the literature has been on the exchange of provisions within a child’s single closest friendship. Still, elementary school children spend most of their school day in a classroom with many peers, making it important to learn in middle childhood whether various classmates and not only a single best friend contribute to the social provisions that children receive in school. Examining the degree to which children receive social provisions from their broader set of classmates provides valuable descriptive information about the exchange of provisions among peers with varying degrees of connectedness within a classroom and, importantly, also allows for the examination of whether provisions received from different sources are linked to children’s feelings of loneliness in school.

Children can have different types of relationships with other children in their classroom. In addition to having one or more reciprocal friendships, children may have unilateral (i.e., unreciprocated) relationships in which the focal child does not name a peer as a friend, but the peer names the child as a friend (unilateral–received relationships), or the focal child names a peer as a friend, but the peer does not name them in return (unilateral–given relationships). Prior research suggests that unilateral relationships occupy a space in between reciprocal friendship and non-friendship in terms of frequency of interaction (Vaughn et al., 2001) and positive engagement with one another (see Newcomb & Bagwell, 1995, for a meta-analysis). Furthermore, although younger (i.e., preschool-aged) children reported that their reciprocal friendship was a better source of social provisions than their unilateral–given friendship, they also reported that both reciprocal and unilateral friendships were better sources of social provisions than non-friendships (Gleason & Hohmann, 2006). We might expect, therefore, that unilateral friendships—in addition to reciprocal friendships—might be sources of provisions that could be protective against loneliness.

Furthermore, even though it seems from prior research that it is unlikely that children receive social provisions as frequently from a non-friend as they do from a friend (e.g., Newcomb & Bagwell, 1995), given the typical self-contained structure of elementary classrooms where children stay together most of the day, a child likely has many interactional opportunities with non-friend classmates that could involve the receipt of provisions. Like provisions exchanged in reciprocal and unilateral friendships, it is possible that provisions exchanged among non-friend classmates could be associated with children’s feelings of loneliness in school.

The Classroom Environment

Beyond a child’s direct experience of receiving provisions from classmates, another potential factor associated with children’s loneliness in school is what we conceptualize as the provision richness of a classroom. Children learning in a classroom where, on average, class members receive help from a higher proportion of their classmates, share companionship with a higher proportion of their classmates, have their contributions validated by a higher proportion of their classmates, and exchange personal information about themselves with a higher proportion of their classmates would likely experience less loneliness than children in a less provision-rich classroom. This broader contextual factor—the provision richness of the classroom—might play a role in loneliness over and above the provisions children individually receive because it could help create a climate of warmth, social opportunity, and exchange. A provision-rich climate might communicate a sense of a caring classroom community that goes beyond the individual-level provisions a child receives.

To our knowledge, research has not assessed the provision richness of classrooms nor its possible association with children’s classroom experiences. However, there is evidence that other aspects of the classroom environment are influential for social, emotional, and academic development (e.g., Barth et al., 2004; Buhs et al., 2015; Gazelle, 2006; see Hamre & Pianta, 2010, Wentzel & Ramani, 2016, and Wentzel, 2022, for reviews). For example, in childhood and early adolescence, the overall level of prosocial behavior and aggression in a classroom have been linked longitudinally with changes in individual students’ prosocial and aggressive behavior (Kuppens et al., 2008; Laninga-Wijnen et al., 2018; Thomas et al., 2011). Also, research on elementary school classroom social structure indicates that, compared to classrooms with more hierarchical social networks, elementary school classrooms with more egalitarian and interconnected social networks at the start of the school year had a higher number of children who were more behaviorally engaged in the academic activities by the end of the year (Cappella et al., 2013). Relatedly, children in classrooms that were more supportive (measured as children’s holistic judgments of classmates’ friendliness and their teacher’s attentiveness) also were concurrently more likely to be behaviorally engaged in the academic content than those in less supportive classrooms (Kilday & Ryan, 2019) and were more likely to become more behaviorally engaged across the school year (Kim & Cappella, 2016). None of the above studies included an examination of classroom environments in relation to feelings of loneliness, although loneliness has been linked with adolescents’ perceptions of their overall school environment in high school (Morin, 2020; Pretty et al., 1994). Given that elementary school children’s lives in school are spent primarily with their classroom peers, we expected that the social provisions environment of the classroom, as measured by children’s reports of provision exchange with each classmate, would be associated with children’s loneliness.

Studies of classroom environment range in their methods for assessing classroom-level variables, with one common approach being to aggregate individual children’s reports about their classmates’ behaviors or other characteristics to create a classroom-level variable (e.g., Boor-Klip et al., 2017; Kuppens et al., 2008; Laninga-Wijnen et al., 2018). We apply the same approach here, aggregating peer reports of social provisions exchange in the classroom to assess the provision richness of the classroom.

Potential Cross-Level Interactions

In addition to a potential main effect association with loneliness, classroom provision richness may also act as a moderator of the association between the provisions individual children receive and their levels of loneliness. Here, contrasting predictions could be made. One possibility is that children who personally receive relatively fewer provisions from their peers could benefit emotionally from the overall positive classroom climate that a provision-rich classroom offers. For these children, classroom provision richness might buffer the effects of receiving fewer provisions on loneliness by enveloping a child in a classroom where peers treat each other with higher levels of warmth and support. Conversely, a provision-rich classroom might exacerbate feelings of loneliness for children who receive relatively fewer provisions compared to their classmates if they realize through social comparison processes that they are having a less rewarding social experience than their peers.

In prior research, other classroom-level characteristics have been found to moderate the association between child-level characteristics and student outcomes. Building on a summer camp study by Wright et al. (1986), research in schools indicated that higher classroom levels of physical aggression attenuated the negative association between child-level aggression and peer acceptance (Boor-Klip et al., 2017), whereas lower classroom levels of relational aggression increased the association between child-level relational aggression and peer rejection (Rohlf et al., 2016). In other research, higher levels of prosocial behavior and academically oriented behaviors at the classroom level were associated with stronger connections between children’s individual prosocial behavior and peer acceptance, and between their individual academic behavior (e.g., doing good work in class) and peer acceptance (Torrente et al., 2014). Together, these findings suggest that moving beyond the individual level of analysis to the group level provides valuable information for understanding children’s lives in school.

Characterizing Other Aspects of Provision-Rich Classrooms

The provision richness of the classroom is likely associated with other sociometric and behavioral characteristics of classroom environments. Given that our provision richness composite score included helping a classmate as one of the provisions, it seems likely that provision-rich classrooms would also be characterized by higher levels of prosocial behavior. Additionally, provision richness, via building good relationships in a classroom, may be negatively associated with the degree of aggression and withdrawn behavior in a classroom.

Although used less frequently to characterize classrooms rather than individual children, classrooms could also be characterized in terms of the average level of peer acceptance and number of reciprocal friendships. Children’s level of peer acceptance and the extent to which children participate in reciprocated friendships are different dimensions of peer social life (Bukowski & Hoza, 1989; Parker & Asher, 1993; Renshaw & Brown, 1993; Vandell & Hembree, 1994). Yet, it is plausible that the provision richness of the classroom would be associated both with the average level of peer acceptance of children in classrooms and with the average level of reciprocated friendships because the exchange of social provisions might have implications for both being well-liked and for forming and maintaining reciprocal relationships.

Gender and Peer Relationships

Gender has been a consistent factor of interest in the peer relations literature, both in studies of students’ same-gender segregation patterns and studies of gender differences in social behavior and relationship characteristics. Children’s preference for and greater amount of time spent with same-gender peers relative to other-gender peers is well-documented (e.g., Hayden-Thomson et al., 1987; Maccoby, 1998; Martin & Fabes, 2001; Mehta & Strough, 2009; Thorne & Luria, 1986), with mixed-gender social groups and friendships occurring less frequently than same-gender groups and friendships (Halim et al., 2021; Lee et al., 2007). For instance, Kovacs et al. (1996) found that fewer than a quarter of children in middle childhood had cross-gender reciprocal friendships based on friendship nominations. Even smaller reciprocal percentages have been found in early adolescent samples (Clark & Ayers, 1992). However, studies using interview methods have found that most children and early adolescents reported having at least one other-gender friendship, although reciprocity was not considered in these cases (Halim et al., 2021; McDougall & Hymel, 2007), leaving open the possibility that for many of these children this reported relationship is unreciprocated. In addition to differences in frequency, children report having different expectations and conceptions of same- and other-gender friendships, suggesting they view these relationships as two separate types (McDougall & Hymel, 2007). Also, with regard to previous studies of gender and friendship, girls’ relationships tend to be more provision-rich than those of boys (see Rose & Rudolph, 2006, for a review), yet gender differences in loneliness are not consistently found (Weeks & Asher, 2012; see Maes et al., 2019, for a meta-analysis). Given the prevalence of gender segregation in peer relationships and documented gender differences in the exchange of provisions in social relationships, we examined provision exchange separately for same- and other-gender peers, and also included child gender in our models. We expected that provisions from same-gender peers would be more common and that they would be more likely to be associated with loneliness than those from other-gender peers.

Goals of the Present Study

To summarize, this study had three primary goals. Our first major goal (Goal 1) was to develop a descriptive picture of the sources of social provisions exchanged in the classroom, given that this study is the first to assess the exchange of provisions among all peers in a classroom. As part of accomplishing this descriptive goal, we examined the gender match between children giving and receiving social provisions and we investigated the degree to which children received provisions from classmates who were not reciprocal friends (Goal 1a). We expected that children would receive more provisions from same-gender than other-gender peers both at the composite level and for individual provisions. We also examined the degree to which children received social provisions in the classroom from: (a) reciprocal friends, (b) unilateral–given and unilateral–received friends, and (c) non-friends (Goal 1b). At the classroom level, we also wanted to characterize classroom-provision richness in terms of other, potentially related, classroom-level features to establish a fuller picture of what these classrooms looked like (Goal 1c). We hypothesized that provision-rich classrooms would also be higher in average peer acceptance and number of reciprocal friends as well as higher in average prosocial behavior level and lower in aggressive and withdrawn behaviors than classrooms that are less provision-rich.

This study’s second major goal (Goal 2) was to examine linkages between social provisions received from (a) reciprocal friends, (b) unilateral–received and unilateral–given friends, and (c) non-friends, and children’s feelings of loneliness in school. We expected that social provisions from reciprocal friends would be associated with lower levels of loneliness, and that receipt of additional provisions from unilateral friends and (to a lesser degree) non-friends might also be associated with lower levels of loneliness, given that these provisions could also help meet children’s social needs.

Our third major goal (Goal 3) was to examine the role played by classroom-level provision richness in children’s feelings of loneliness and to learn whether classroom-level provision richness moderates the effect of child-level provision receipt on loneliness. We hypothesized that classroom-level provision richness would also be associated with loneliness as a main effect (Goal 3a) and might serve as a moderator of child-level provision receipt (Goal 3b). We considered alternative hypotheses concerning the direction of the classroom-level moderation effect, specifically a buffering effect in which receiving fewer provisions in a more provision-rich classroom is associated with lower loneliness than in a less provision-rich classroom, and an exacerbating effect, in which receiving fewer provisions would be associated with greater feelings of loneliness in a more provision-rich classroom than in classrooms that are less provision-rich.

Method

Participants

Participants were from 38 third- through fifth-grade classrooms in two Midwest U.S. public elementary schools outside a metropolitan area. Participants were 98% of the children in those grades. Of the 15 children who did not participate, one did not have parental approval, seven had just moved to the school or were primarily homeschooled, and seven had significant reading problems or other circumstances that limited their participation. The final sample was 998 children (468 girls, 530 boys), of whom 103 girls and 113 boys were in nine third-grade classrooms, 159 girls and 182 boys were in 13 fourth-grade classrooms, and 206 girls and 235 boys were in 16 fifth-grade classrooms. The average class size was 26 children (range = 22–30). Regarding race/ethnicity, 88.5% of the sample was White, 2.2% was Black, 1.8% was Middle Eastern, 3.5% was Asian, 1% was Latinx, and 3% was other races/ethnicities. Census information for families in the school district indicated that approximately 12% of families were at or below two times the federal poverty threshold for a family of four, and 64% of families had incomes that were between approximately two and five times this threshold, and the remaining 24% of families had incomes approximately five or more times the threshold. The participating schools were the two intermediate schools (Grades 3–5) in the school district. The school district was asked to participate because they had participated in a prior study and the community was socioeconomically diverse.

Procedure

Consistent with U.S. federal guidelines [45 Code of Federal Regulations 46.116(d)], an IRB-approved waiver of written parental consent procedure was used. This procedure was also approved by the school administrators and is consistent with those used in prior research on topics such as friendship, interpersonal goals, loneliness, adolescent coping methods, and tobacco use (e.g., Ebesutani et al., 2012; Mayeux et al., 2008; McDonald & Asher, 2018; Rose & Asher, 1999; Siqueira et al., 2000). As part of the approved procedure, schools sent a letter to parents approximately 2 weeks before the start of data collection. The letter described the project and asked parents to contact the school if they did not want their child to participate or if they would like more information. With regard to child assent for participation, children were told at the start of each session that they did not have to complete the questionnaires if they preferred not to and that they could skip items they did not want to answer. Children who were absent on administration days had the opportunity to complete the measures on make-up days (see the Data Analysis section for missing data information).

Data for this study were drawn from a larger research project involving three weekly data-collection sessions of 50 min or less. The measures used in this study were administered in Session 2 (peer behavioral nominations and classroom social provisions) and Session 3 (loneliness, friendship nominations, and peer acceptance ratings). Data collection sessions were conducted by four female graduate research assistants, with a different research assistant assigned to a class for each session. Research assistants followed a script to ensure consistency. Materials and data analysis code for this study are available by emailing the corresponding author.

Measures

Loneliness

The Loneliness in Context questionnaire for children (Asher et al., 2023) was used to measure feelings of loneliness in four school contexts (i.e., classroom, playground, lunchroom, and physical education), with three items per context, using these question stems: “Are you lonely in [context]?”; “Is [context] a lonely place for you?”; and “Do you feel sad and alone in [context]?”. The 12 loneliness items were interspersed with 15 neutral filler items about school activities and interests to balance out the potentially emotionally valanced loneliness questions. Children responded to each item on a 5-point Likert scale (1 = never, 5 = always). Responses to the 12 loneliness items were summed to create an overall loneliness score, which could range from 12 to 60. Internal consistency reliability for the overall score was α = .94.

A key feature of the Loneliness in Context measure is that it focuses exclusively on feelings of loneliness and not on potential causes of loneliness such as the receipt of provisions (such as help and support) from social relationships, having friends, being liked by other children, or feeling socially competent. Such diverse item content is included in most widely used measures of loneliness (see Weeks & Asher, 2012, for a review). Using a highly focused measure of loneliness is crucial in this research to avoid overlap between measures assessing key variables of interest (e.g., the receipt of provisions from social relationships) and measures of loneliness. For a fuller discussion of the importance of using highly focused loneliness measures when studying the causes of loneliness, see Weeks and Asher (2012), and for a broader discussion of how overlapping content in the measurement of the independent variable and the dependent variable is highly problematic in various lines of inquiry, see Nicholls et al. (1982).

Although the Loneliness in Context measure is a newer measure, it builds on earlier research using a subset of a few “pure” loneliness items drawn from the widely used loneliness and social dissatisfaction measures (Asher et al., 1984; Asher & Wheeler, 1985) to assess children’s feelings of loneliness in school without overlapping or confounding content (e.g., Kochenderfer-Ladd & Wardrop, 2001; Ladd et al., 1996; Ladd & Troop-Gordon, 2003; Lavallee & Parker, 2009; Parker & Asher, 1993). Such brief “pure” loneliness measures have good internal consistency reliability and have been related in expected directions with peer acceptance (e.g., Ladd & Troop-Gordon, 2003; Parker & Asher, 1993), friendship quality (e.g., Lavallee & Parker, 2009; Parker & Asher, 1993), and school liking and avoidance (e.g., Ladd et al., 1996), supporting their use as valid assessments of children’s feelings of loneliness in school.

Social Provisions

To assess the social provisions children received from their classmates, children completed a classroom-based peer-report social provisions measure in which they responded to items about which children in their class to whom they provided companionship, help, validation, and opportunities for self-disclosure. To our knowledge, this study is the first to assess the exchange of provisions in a classroom in this way. However, peer nomination methodologies are regularly used in middle childhood to study children’s behavioral characteristics (e.g., Coie et al., 1982; Crick & Grotpeter, 1995; Kuppens et al., 2008; see Bierman, 2004, and Cillessen & Marks, 2017, for reviews), suggesting that nomination-type measures are appropriate for children of this age.

For each social provision item, children were given a separate class roster with a description of the specific provision at the top of the page. The research assistant read the item aloud and asked children to circle the names of the classmates with whom they engaged in that provision-related behavior. The 12 provision items were based on items from Parker and Asher’s (1993) Friendship Quality Questionnaire. Prior to analysis, we made the decision to remove four of these 12 items (one item per provision) given concerns that they did not fit as well conceptually with the other items on their respective provision subscale (see Supplementary Materials Table S1 for items that were removed and for the specific rationale for removing each item). The final set of provision questions included eight items (two per provision); these items are shown in Table 1 along with descriptive and internal consistency reliability information.

Table 1.

Items, Internal Consistency Reliability, and Descriptive Statistics for Social Provisions Received Proportion Scores

Provision Category	Items	SG M (SD)	SG Range	OG M (SD)	OG Range
Companionship SG α = .74 OG α = .59	Who do you spend time with at lunch or recess?	.49 (.22)	0–1	.04 (.08)	0–.54
	Who do you often pick as partners for things?	.34 (.18)	0–1	.01 (.04)	0–.54
Validation SG α = .74 OG α = .71	Who do you tell they’re good at things?	.43 (.21)	0–1	.12 (.12)	0–.83
	Who do you try to make feel important and special?	.34 (.19)	0–1	.09 (.10)	0–.62
Help SG α = .71 OG α = .72	Who do you do favors for?	.41 (.19)	0–1	.11 (.12)	0–.71
	Who do you give help and advice when they are having trouble figuring something out?	.34 (.18)	0–.89	.12 (.11)	0–.77
Opportunity for self-disclosure SG α = .77 OG α = .67	Who tells you private things about themselves?	.18 (.15)	0–.83	.03 (.06)	0–.36
	Who talks to you when they’re sad or mad about something that happened to them?	.23 (.17)	0–1	.03 (.06)	0–.40
Composite Proportion Score		.35 (.14)	0–.81	.07 (.06)	0–.39

Note. SG = same gender, OG = other gender.

To examine provisions that individual children received, we calculated two types of scores (described in the following sections) to be used in different types of analyses. For both types of scores, they were calculated separately for same- and other-gender peers, given pervasive gender segregation in middle childhood.

Proportion of Same- and Other-Gender Peers From Whom Provisions Nominations Were Received.

First, we calculated same- and other-gender provisions received proportion scores separately for each of the eight provisions items, indicating the proportion of same- or other-gender classmates who circled the target child’s name for that item. For instance, a girl who received nominations on the help item “Who do you do favors for?” from three of the 12 girls in her class and one of the 10 boys in her class would have a same-gender provisions received proportion score of .25 and an other-gender provisions received proportion score of .10 for that item. Next, the overall provisions received proportion scores (same- and other-gender) for each child were calculated, averaging the proportion of same- or other-gender classmates who nominated the child across all eight provisions items.

Sum of Same- and Other-Gender Peers From Whom Provisions Nominations Were Received.

We also calculated the totals of same- and other-gender provision nominations received (sum scores) separately for each of the eight items, using the sum (total number) of same- or other-gender classmates who circled the target child’s name for that item. For example, a girl who received nominations from three of the 12 girls in her class and one of the 10 boys in her class on the help question “Who do you do favors for?” would have a same-gender sum nomination score of 3 and an other-gender sum nomination score of 1 for that item. A child’s overall sum nomination scores (same- and other-gender) were then calculated: the sum of the peer nomination scores across the eight items, separately for same- and other-gender peers.

Sum scores were calculated in addition to the proportion scores for two reasons. First, the overall quantity of provisions received might be particularly relevant for the degree of loneliness children experience. Second, in certain analyses we focus specifically on subgroups of peers (i.e., reciprocal friends, unilateral friends, and non-friends, for analyses addressing Goals 1b and 2). In those cases, determining the appropriate denominator for proportion scores was unclear, and interpretations of comparisons would be more straightforward with sum scores.

Same- and Other-Gender Classroom Provision Richness.

Next, to assess a classroom’s provision richness, classroom-level provision richness scores were calculated as the average of the overall provisions-received proportion scores of the students in each classroom. As with the other provisions scores, classroom provision richness scores were calculated separately for provisions received from same- and other-gender peers. For example, a classroom same-gender provision richness score of .25 indicates that children in that room, on average, received provisions from 25% of their same-gender classmates. This aggregation method is consistent with those used in recent research to obtain classroom-level metrics of prosocial behavior and aggression (e.g., Boor-Klip et al., 2017; Kuppens et al., 2008; Laninga-Wijnen et al., 2018). As conceptualized by Stapleton et al. (2016), this would represent a configural classroom-level construct (also referred to in the literature as a contextual or formative construct), representing “the aggregate of the disparate characteristics of individuals within a cluster” (p. 488).

Friendship Nominations

Children completed both unlimited and limited friendship nomination measures. Children first were given a roster of their classmates’ names and were asked to circle the names of all of their “very best friends” in the class. Then, on a second roster, children were asked to circle the names of their “three best friends.” Our goal was to identify children who nominated each other as friends (i.e., had reciprocal nominations). The unlimited friendship nominations measure provides a more inclusive identification of friendships than the limited nomination measure because children are not limited in their nominations and can select all classmates they consider to be a friend (see Bagwell & Schmidt, 2011, and Bagwell et al., 2021, for fuller discussions of the rationale for using an unlimited friendship nomination measure).

The more inclusive procedure was appropriate for the current research questions, and therefore the unlimited friendship nominations measure was used. In the present study, children on average made 4.51 same-gender (SD = 2.80, range = 0–17) and 0.40 other-gender (SD = 1.23, range = 0–16) nominations, and based on these nominations and the nominations they received, children had an average of 2.62 same-gender (SD = 1.83, range = 0–11) and 0.10 other-gender (SD = 0.41, range = 0–5) reciprocal friendships.

Peer Acceptance

To assess children’s level of peer acceptance in the classroom, children provided sociometric ratings on a 5-point scale (1 = don’t like to, 5 = like to a lot) of how much they liked to play with each classmate (i.e., “How much do you like to play with this person at school?”). A rating-scale sociometric measure (e.g., Kochel et al., 2017; McDonald & Asher, 2018; Schofield & Whitley, 1983; Singleton & Asher, 1977) was used because providing information about how every child feels about each of their classmates affords complete information for computing each child’s peer acceptance score (Bierman, 2004). Furthermore, rating-scale sociometric measures are likely to yield peer acceptance scores that are more statistically independent from friendship nominations than would “like most”/“like least” nomination measures of peer acceptance, since like-most nominations and friendship nominations both involve choosing classmates who children presumably like the most (see Asher & Weeks, 2018, for a fuller discussion of this issue). Rating-scale sociometric measures are commonly used in early and middle childhood and adolescence (e.g., Cassidy & Asher, 1992; Hundley & Cohen, 1999; Ladd, 1981; Nangle et al., 2004; Oden & Asher, 1977) and yield fairly stable peer acceptance scores over time, even among younger children (e.g., Asher et al., 1979; Kingery et al., 2011). As with social provision scores, we examined acceptance scores separately for same-gender and other-gender peers. Same-gender and other-gender peer acceptance scores were calculated for each child as the average liking rating they received from same- or other-gender classmates. These scores were in turn used to calculate classroom-level scores, which were the classroom averages of the same-gender acceptance ratings and the other-gender acceptance ratings children received from classmates. In contrast with many previous studies using a rating-scale sociometric measure to assess peer acceptance, unstandardized acceptance ratings were used rather than standardizing children’s peer acceptance scores within classrooms because we were interested in variability across classrooms in average peer acceptance ratings.

Peer Assessment of Social Behavior

Children completed a 12-item peer assessment of their classmates’ prosocial behaviors (“cooperative,” “helpful,” “friendly,” “honest”), withdrawn behaviors (“afraid to join in a group,” “likes to be alone a lot,” “shy,” “easy to push around”), and aggressive behaviors (“mean,” “starts fights,” “hits, pushes, or kicks,” “gets mad easily”; Rockhill & Asher, 1992). Peer assessments of classmates’ social behavior are commonly used in elementary school (e.g., Coie et al., 1982; Crick & Grotpeter, 1995; Kuppens et al., 2008) and provide reliable, stable, and valid indices of children’s social behavior in the classroom (e.g., Masten et al., 1985; Serbin et al., 1987; see Bierman, 2004, and Cillessen & Marks, 2017, for reviews). For each of the behavioral nomination items, children were presented with a separate class roster with the behavior at the top of the page and were asked to circle the names of classmates who fit that particular descriptor. No limits were placed on the number of names children could circle.

Each child received same- and other-gender scores for prosocial behavior, withdrawn behavior, and aggressive behavior. Using the four items of a particular subscale, subscale scores were calculated as the average proportion score of same- or other-gender classmates who nominated the child as displaying each of the behaviors. The internal consistency reliability was good for the prosocial (α_same-gender = .93; α_other-gender = .91), aggressive (α_same-gender = .94; α_other-gender = .94), and withdrawn (α_same-gender = .77; α_other-gender = .75) behavior subscales. Prior research with the measure has found comparable internal consistency reliability (Chung & Asher, 1996). Classroom-level same- and other-gender behavior scores (see Table 2 for descriptive information) were calculated in the same manner as the classroom-level provision richness scores described above.

Table 2.

Descriptive Statistics for Classroom-Level Variables

	M	SD	Range	Correlation with SG Provision Richness	Correlation with OG Provision Richness
Provision Richness SG	.34	.06	.23–.48
Provision Richness OG	.07	.04	.01–.19
Prosocial Behavior SG	.53	.06	.43–.63	.52***†
Prosocial Behavior OG	.31	.07	.14–.47		.58***†
Aggressive Behavior SG	.17	.03	.11–.23	−.06
Aggressive Behavior OG	.16	.04	.09–.23		.03
Withdrawn Behavior SG	.12	.03	.06–.20	−.08
Withdrawn Behavior OG	.11	.02	.05–.16		.13
Peer Acceptance SG	3.44	.25	2.86–3.91	.45**
Peer Acceptance OG	1.72	.32	1.12–2.32		.63***
Reciprocated Friends SG	2.61	.61	1.73–4.14	.10
Reciprocated Friends OG	.10	.15	0–.67
Loneliness	17.38	2.15	14.18–22.67

Note. Classroom n = 38. SG = same-gender; OG = other-gender. Means are the average of the same-gender or other-gender child-level provision received proportion scores within a classroom. Due to the small number of classrooms in which reciprocal other-gender friendships occurred (n = 18), the correlation between other-gender social provision richness and other-gender reciprocal friendships was not examined. Although we used the withdrawn subscale from Rockhill and Asher (1992), this subscale includes a wide range of behaviors that could be characterized as withdrawn, submissive, or shy. We tested whether any of these behaviors alone correlated with provision richness, but none did. Furthermore, removing items from the withdrawn composite did not improve the internal consistency reliability of the scale, so we kept this composite despite the variation in behaviors that these items reference.

^†These correlations remained significant, but were slightly weaker (r_same-gender = .45, p = .005; r_other-gender = .52, p < .001) when the two help provisions items were removed from the provisions composite due to conceptual overlap of the two help items with the assessment of prosocial behavior.

^*p < .05

^**p < .01

^***p < .001.

Data Analysis Plan

Missing Data

Missing data were minimal. All 998 participants had “received” scores (i.e., the proportion of peers from whom children received social provisions, the average peer acceptance rating children received from peers, and the proportion of peers who nominated children as prosocial, withdrawn, or aggressive). For the self-report measures used in this study, 17 children (0.02%) were missing only the friendship nominations-given measure, seven children (0.01%) were missing only the loneliness measure, two children (0.003%) were missing the friendship nominations-given and loneliness measures, one child (0.001%) was missing the friendship nominations-given and provisions-given measures, and two children (0.003%) were missing all three measures. Item-level missing data on the loneliness measure ranged from 0.7%–1.4%. Patterns of missingness were examined and no concerning patterns were noted.

Given that such a small degree of missingness has a negligible effect on estimates (Newman, 2014; Schafer & Graham, 2002), children with missing data were excluded from relevant analyses via pairwise deletion. The children missing one or more measures did not significantly differ on study measures from the children who had complete data. The final sample size for each analysis ranged from 970 to 998 depending on whether loneliness and/or friendship measures were included in a particular analysis.

Statistical Analyses

Given our interest in both child-level and classroom-level social provision effects, we used multilevel modeling to address our goals, apart from analyses focusing only on the classroom level of analysis (Goal 1b). All multilevel models were estimated in SAS PROC MIXED (version 9.4) using maximum likelihood estimation. The substantive conclusions of these analyses are identical when using restricted maximum likelihood, and thus results are reported with maximum likelihood estimation to allow for comparison of nested models using the deviance statistic (−2 log likelihood; e.g., Hox, 2010). Multilevel regression models predicting loneliness were estimated using a “bottom-up” approach where predictors are added to the models in steps and improvement in prediction is assessed via significance tests of individual parameters and χ² difference tests (using the −2 log likelihood, or “deviance” statistic) between successive models (e.g., Hox, 2010). If successive models are indeed a better fit to the data, the deviance statistic should significantly decrease with the addition of predictors, indicated by a statistically significant χ² difference test. The specific analyses for each goal are described below.

To address Goal 1a regarding gender differentiated receipt of social provisions, we estimated a three-level multilevel model with provisions from peers by gender (same- or other-gender) and provision by type (companionship, validation, help, and intimate exchange) as repeated measures within children at Level 1, children within classrooms at Level 2, and classrooms at Level 3. To address Goal 1b regarding from whom in their classroom children received social provisions, we examined descriptive statistics for the sum provisions received from different relationship partners (reciprocal, unilateral–given and unilateral–received, and non-friends). To address Goal 1c regarding the classroom-level correlates of provision richness, we examined zero-order correlations among provision richness, peer acceptance, number of reciprocal friends, and prosocial, aggressive, and withdrawn behaviors, all at the classroom level. Note that, with a sample of 38 classrooms, these correlational analyses addressing Goal 1c are powered (at the 80% level with a p threshold of .05) to detect correlations of .44 or greater, so more modest classroom-level associations would not reach the threshold of statistical significance in the present study.

To address Goal 2, examining the degree to which provisions from different sources contribute to loneliness, we estimated a two-level multilevel model, within children (Level 1) nested within classrooms (Level 2). The sum provisions received scores from each relationship source (i.e., reciprocal, unilateral–given and unilateral–received, and non-friends) were entered as predictors of loneliness in successive steps. Grade and gender were included as control variables and child-level provisions variables were group-mean centered within class. In these analyses, children who did not have a particular relationship type were given a score of zero on the provision variables for that relationship type. This decision adds ambiguity to the meaning of a zero score because it could reflect either (a) that a child had that type of relationship but did not receive provisions from that peer, or (b) that a child did not have that type of relationship and therefore could not receive any provisions from that source. Our decision to consider both cases to be zero scores was based on two reasons. First, even though it is true that there are two different pathways to zero, in both cases, children are not receiving provisions from that relationship type. Second, were we to limit the analyses to only those children who had at least one relationship of each of the eight types (i.e., same- and other-gender reciprocal, unilateral–given, unilateral–received, and non-friends) it would severely limit the sample size and relatedly, would yield findings that are not generalizable to children more broadly. This smaller group of children would have a relationship profile that is very different from nearly all children (i.e., these children have all eight relationship types, which only three out of 998 children in this sample had; see Table 3 and Table 4 for frequency of relationship types and for mean sum scores from each type).

Table 3.

Descriptive Statistics for Relationship Types from Unlimited Friendship Nominations

	n at zero	M	Range	SD
Reciprocal SG	101	2.62	0–11	1.83
Reciprocal OG	907	0.10	0–5	0.41
Unilateral–Given SG	353	1.83	0–14	2.21
Unilateral–Given OG	841	0.29	0–14	1.04
Unilateral–Received SG	259	1.86	0–10	1.73
Unilateral–Received OG	789	0.29	0–5	0.65
Non-friend SG	33	5.93	0–16	3.02
Non-friend OG	3	12.04	0–19	2.57

Note. n = 977 for all friend types except unilateral–received (n = 998) due to missing friendship nomination measures for 21 children. N at zero refers to the number of participants who had zero relationships of that type.

SG = same gender; OG = other gender.

Table 4.

Sum Provisions Received by Relationship Types for Children Who Have At Least One Relationship of that Type

	n	M	SD	Range
Reciprocal SG	876	15.35	9.11	0–59
Reciprocal OG	70	7.56	5.78	0–31
Unilateral-Given SG	624	6.02	5.14	0–32
Unilateral-Given OG	136	1.99	3.00	0–23
Unilateral-Received SG	739	9.07	7.25	0–39
Unilateral-Received OG	209	3.78	3.31	0–20
Non-friend SG	944	7.97	6.34	0–36
Non-friend OG	974	5.20	4.74	0–30

Note. Means reflect the average sum provisions score from that relationship type. These descriptive statistics only include children who had at least one relationship of that relationship type based on the unlimited friendship nomination measure.

SG = same gender; OG = other gender.

To address Goals 3a and 3b, examining main effects of child-level provisions received, classroom provision richness, and their interaction on feelings of loneliness, two-level multilevel models (with children, Level 1, nested within classrooms, Level 2) were estimated. Child-level provisions received proportion scores from same- and other-gender peers, classroom-level same- and other-gender provision richness, and the cross-level interaction were entered in successive steps as predictors of loneliness. For these analyses, provisions received from classmates (same- and other-gender separately) were used as composite measures of child-level provisions received for the sake of parsimony. Grade and gender were included as control variables.

Following Enders and Tofighi’s (2007) recommendations for situations in which main effects and cross-level interactions are of substantive interest, child-level (Level 1) provisions received scores were centered at the group mean (i.e., within each classroom), and classroom-level (Level 2) provisions received scores were centered at the grand mean (i.e., across all classrooms). Group mean centering Level 1 (child-level) predictor variables allows for more accurate estimates of within-classroom effects because between-class variability is removed, and it is necessary when estimating cross-level interactions (Enders & Tofighi, 2007). Grand mean centering Level 2 (classroom-level) predictors facilitates interpretation of cross-level interaction effects but does not affect substantive conclusions.

Results

Correlations among study variables are presented in Table 5. Mean-level grade and gender differences in child-level study variables are presented in Appendix Tables A1–A3. Intraclass correlation coefficients for the provisions received variables are in Table 6 and descriptive statistics for provisions received and loneliness for each class are in Appendix Table A4.

Table 5.

Correlations between Loneliness and Social Provisions Received from Same- and Other-Gender Peers

	1	2	3	4	5	6	7	8	9	10
1. Loneliness
2. Social Provisions Overall SG	−.17***
3. Companionship SG	−.22***	.81***
4. Help SG	−.14***	.87***	.57***
5. Validation SG	−.11***	.89***	.63***	.72***
6. Self-disclosure SG	−.09**	.80***	.47***	.67***	.63***
7. Social Provisions Overall OG	.01	.24***	.14***	.19***	.28***	.19***
8. Companionship OG	.03	.15***	.11***	.10**	.18***	.13***	.74***
9. Help OG	.01	.21***	.11***	.18***	.24***	.18***	.88***	.53***
10. Validation OG	−.02	.23***	.17***	.18***	.29***	.14***	.85***	.50***	.62***
11. Self-disclosure OG	.03	.16***	.06	.15***	.17***	.19***	.79***	.63***	.61***	.53***

Note. SG = same gender, OG = other gender.

Social provision scores are provisions received proportion scores.

^*p < .05

^**p < .01

^***p < .001.

Table 6.

Intraclass Correlation Coefficients for Social Provision Received Variables

	Intraclass Correlation Coefficient (ICC)
	Same Gender	Other Gender
Companionship	.10	.34
Intimate Exchange	.13	.31
Validation	.18	.46
Help	.18	.31
Social Provisions Overall	.14	.41

Note. These ICCs reflect social provisions from all classmates.

Goal 1: Descriptive Examination of the Classroom Provision Environment

Goal 1a: Receipt of Provisions From Same- Versus Other-Gender Peers

To examine whether children receive more provisions from same- versus other-gender peers and whether the proportion of classmates from whom children received provisions varied across the four provision types, we estimated a three-level model with provision type (i.e., companionship, help, validation, and opportunity for self-disclosure) and gender of peer (i.e., same- vs. other-gender) nested within child, and children nested within classrooms (see Table 6 for intraclass correlation coefficients [ICCs] and Table 7 for multilevel parameter estimates with each provision type as the reference group). Consistent with our expectations, children received provisions from a much higher proportion of their same-gender peers (M = .35, SD = .14) than their other-gender peers (M = .07, SD = .06). Main effects also emerged for provision type with all provisions’ proportion scores significantly differing from each other except for help and validation. These main effects were qualified by significant gender-of-peer by provision-type interactions. For same-gender peers, children received companionship from the highest proportion of peers (M = .41, SD = .18), which was significantly higher than validation (M = .38, SD = .18) and help (M = .38, SD = .16; validation and help did not significantly differ from each other, see Table 7). Children received opportunities for self-disclosure from a significantly smaller proportion of their same-gender peers (M = .21, SD = .14) than for all other types of provisions. For other-gender peers, the proportion of other-gender peers from whom children received help (M = .12, SD = .10) and validation (M = .11, SD = .10) did not significantly differ (Table 7). Children received opportunities for self-disclosure and companionship from very few other-gender peers (M = .03, SD = .05 and M = .03, SD = .05, respectively) and these proportions did not significantly differ from each other, but did differ from help and validation.

Table 7.

Parameter Estimates from Multilevel Model Examining Provisions from Same- and Other-Gender Peers by Provision Type

	Companionship as Reference Group	Help as Reference Group	Validation as Reference Group	Intimate Exchange as Reference Group
	Estimate (se)	Estimate (se)	Estimate (se)	Estimate (se)
Parameter Estimates with Gender of Peer Coded as 0 = Same-Gender, 1 = Other-Gender
Intercept	0.42 (0.008)***	0.38 (0.008)***	0.38 (0.008)***	0.21 (0.008)***
Same- vs. Other-Gender Peer	−0.39 (0.005)***	−0.26 (0.005)***	−0.27 (0.005)***	−0.18 (0.005)***
Reference Group vs. Companionship		0.04 (0.005)***	0.03 (0.005)***	0.21 (0.005)***
Reference Group vs. Help	−0.04 (0.005)***		−0.01 (0.005)	0.17 (0.005)***
Reference Group vs. Validation	−0.03 (0.005)***	0.01 (0.005)		0.17 (0.005)***
Reference Group vs. Intimate Exchange	−0.21 (0.005)***	−0.17 (0.005)***	−0.17 (0.005)***
Peer Gender x RG vs. Companionship		−0.13 (0.007)***	−0.11 (0.007)***	−0.21 (0.007)***
Peer Gender x RG vs. Help	0.13 (0.007)***		0.01 (0.007)*	−0.08 (0.007)***
Peer Gender x RG vs. Validation	0.11 (0.007)***	−0.01 (0.007)*		−0.09 (0.007)***
Peer Gender x RG vs. Intimate Exchange	0.21 (0.007)***	0.08 (0.007)***	0.09 (0.007)***
Parameter Estimates with Gender of Peer Coded as 0 = Other-Gender, 1 = Same-Gender
Intercept	0.03 (0.008)***	0.12 (0.008)***	0.11 (0.008)***	0.03 (0.008)***
Other-Gender vs. Same-Gender Peer	0.39 (0.005)***	0.26 (0.005)***	0.27 (0.005)***	0.18 (0.005)***
Reference Group vs. Companionship		−0.09 (0.005)***	−0.08 (0.005)***	−0.002 (0.005)
Reference Group vs. Help	0.09 (0.005)***		0.009 (0.005)	0.09 (0.005)***
Reference Group vs. Validation	0.08 (0.005)***	−0.009 (0.005)		0.08 (0.005)***
Reference Group vs. Intimate Exchange	0.002 (0.005)	−0.09 (0.005)***	−0.08 (0.005)***
Peer Gender x RG vs. Companionship		0.13 (0.007)***	0.11 (0.007)***	0.21 (0.007)***
Peer Gender x RG vs. Help	−0.13 (0.007)***		−0.01 (0.007)*	0.08 (0.007)***
Peer Gender x RG vs. Validation	−0.11 (0.007)***	0.01 (0.007)*		0.09 (0.007)***
Peer Gender x RG vs. Intimate Exchange	−0.21 (0.007)***	−0.08 (0.007)***	−0.09 (0.007)***
τ00(2)	0.004 (0.0002)***	0.004 (0.0002)***	0.004 (0.0002)***	0.004 (0.0002)***
τ00(3)	0.002 (0.0005)***	0.002 (0.0005)***	0.002 (0.0005)***	0.002 (0.0005)***
σ 2	0.012 (0.0002)***	0.012 (0.0002)***	0.012 (0.0002)***	0.012 (0.0002)***

Note. Models were estimated in SAS 9.3 using PROC MIXED with maximum likelihood estimation (substantive conclusions are identical using restricted maximum likelihood estimation).

RG = reference group, se = standard error. τ₀₀⁽³⁾ = estimate of variance at level 3 (classroom level); τ₀₀⁽²⁾ = estimate of variance at level 2 (child level); σ² = estimate of variance at level 1 (provision level).

Estimates of variance at each level are identical regardless of gender of peer coding.

^*p < .05

^**p < .01

^***p < .001; to correct for multiple comparisons, provision-type comparisons are considered significantly different from zero if p < .001 (.05/48 comparisons).

Finally, we also examined mean differences between classrooms (n = 38). At the classroom level, the average same-gender provision richness score was significantly higher (M = .34, SD = .06) than the average other-gender provision richness score (M = .07, SD = .04), t(37) = 34.63, p < .001, d = 5.62.

Goal 1b: Descriptive Picture of Relationship Sources of Social Provisions

To address the question of from whom children received social provisions, we used the unlimited friendship nomination data to identify children’s (a) reciprocal friends, (b) unilateral–given friends (target child’s nomination of friendship is not returned), (c) unilateral–received friends (target child received but did not provide a reciprocal nomination of friendship), and (d) non-friends (neither child nominated the other as a friend). In approximately 4% of same- and other-gender cases, the nature of the relationship between children could not be identified due to missing friendship nomination data, so the sources of those provision nominations were classified as “unknown.”

Although children received many of their provision nominations from reciprocal friends, more provision nominations were received from classmates with whom a child did not have a reciprocal relationship. For the sum number of provision nominations received from same-gender classmates, 39.59% of the provision nominations contributing to this score were, on average, from reciprocal friends, with 13.42% coming from unilateral–given friends, 18.76% from unilateral–received friends, and 24.51% from non-friends. The additional 3.72% were from classmates where the relationship status was unknown. Regarding the sum provision nominations received from other-gender peers, 3.58% were from reciprocal friends, 3.41% were from unilateral–given friends, 8.19% were from unilateral–received friends, and 80.49% were from non-friends (4.33% relationship status unknown).

Goal 1c: Classroom-Level Correlates of Provision Richness

We also examined the characteristics of provision-rich classrooms in relation to classroom-level sociometric indices (i.e., peer acceptance and number of reciprocal friends) and behavioral style (i.e., prosocial behavior, aggression, and withdrawal; see Table 2 for descriptive statistics). Correlational analyses indicated that classrooms higher in provision richness were also significantly higher in the average level of peer acceptance and peer-reported prosocial behavior (see Table 2). The correlation between classroom provision richness and prosocial behavior remained significant even when the help provision was removed from the classroom provision richness measure to limit potential conceptual overlap between the measures (see Note in Table 2). Classroom provision richness was not significantly associated with the average number of same-gender reciprocal friends in the classroom. We did not examine the correlation between other-gender provision richness and other-gender reciprocal friendship because 20 of the 38 classrooms had zero reciprocal other-gender friendships. Finally, regardless of whether the behavioral variables were based on same- or other-gender classmates’ nominations, classroom provision richness was not significantly associated with classroom-level aggression or withdrawal.

Goal 2: Provisions by Relationship Source and Their Association With Loneliness

Turning to the association of provisions with loneliness, we next conducted a multilevel regression analysis to examine whether sum provisions received from unilateral friendships or non-friends explained variation in loneliness over and above that explained by sum provisions received from reciprocal friends, controlling for child gender (Level 1; 0 = girls, 1 = boys) and grade (Level 2; 0 = 3^rd grade, 1 = 4^th grade, 2 = 5^th grade). Parameter estimates, fit statistics, and variance components for all steps are presented in Table 8 and correlations among variables included in this analysis are in Table 9.

Table 8.

Parameter Estimates from Multilevel Regressions Predicting Children’s Loneliness from Provisions Received from Reciprocal Friends, Unilateral Friends, and Non-Friend Classmates

	Intercept-Only Model	Provisions Received from Reciprocal Friends	Provisions Received from Unilateral Friends	Provisions Received from Other Classmates
	Estimate (se)	Estimate (se)	Estimate (se)	Estimate (se)
Intercept	17.36 (0.33)***	19.40 (0.60)***	19.52 (0.60)***	19.59 (0.60)***
Child Gender (Child Level)		−0.84 (0.50)	−1.04 (0.50)*	−1.17 (0.52)*
Grade (Classroom Level)		−1.31 (0.37)**	−1.32 (0.37)**	−1.32 (0.37)**
Child-Level Sum Provisions
Received from Reciprocal Friends SG		−0.14 (0.03)***	−0.14 (0.03)***	−0.15 (0.03)***
Child-Level Sum Provisions
Received from Reciprocal Friends OG		−0.03 (0.11)	−0.06 (0.11)	−0.06 (0.11)
Child-Level Sum Provisions
Received from Unilateral–Received Friends SG			−0.09 (0.04)*	−0.09 (0.04)*
Child-Level Sum Provisions
Received from Unilateral–Received Friends OG			−0.02 (0.13)	−0.01 (0.13)
Child-Level Sum Provisions
Received from Unilateral–Given Friends SG			−0.09 (0.06)	−0.11 (0.06)
Child-Level Sum Provisions
Received from Unilateral–Given Friends OG			0.36 (0.20)	0.34 (0.20)
Child-Level Sum Provisions
Received from Non-Friend Classmates SG				−0.05 (0.05)
Child-Level Sum Provisions
Received from Non-Friend Classmates OG				−0.04 (0.07)
τ00	1.83 (1.00)*	0.86 (0.75)	0.88 (0.74)	0.87 (0.74)
σ 2	62.20 (2.86)***	60.25 (2.77)***	59.61 (2.74)***	59.54 (2.73)***
AIC	6905.10	6872.20	6870.00	6872.70
BIC	6910.10	6883.70	6888.10	6894.00
Deviance (−2 log likelihood)	6899.10	6858.20	6848.00	6846.70
Δχ2		40.90***	10.20*	1.30

Note. Models were estimated in SAS 9.3 using PROC MIXED with maximum likelihood estimation (substantive conclusions are identical using restricted maximum likelihood estimation).

Child gender is coded 0 = girls, 1 = boys; grade is coded 0 = 3^rd grade, 1 = 4^th grade, 2 = 5^th grade; SG = same-gender, OG = other-gender; se = standard error. Child-level provisions scores are mean-centered within classrooms. The intercept was modeled as a random effect, all others are fixed effects; τ₀₀ = estimate of variance at Level 2 (classroom level); σ² = estimate of variance at Level 1 (child level). AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion.

^*p < .05

^**p < .01

^***p < .001.

Table 9.

Correlations Among Variables Included in Multilevel Regressions Predicting Children’s Loneliness from Provisions Received from Reciprocal Friends, Unilateral Friends, and Non-Friend Classmates (Goal 2)

	1.	2.	3.	4.	5.	6.	7.	8.	9.	10.
1. Grade
2. Gender	.01
3. Child-Level Sum Provisions Received from Reciprocal Friends SG	.00	.05
4. Child-Level Sum Provisions Received from Reciprocal Friends OG	.00	−.01	.02
5. Child-Level Sum Provisions Received from Unilateral–Given Friends SG	.00	−.11***	−.13**	−.04
6. Child-Level Sum Provisions Received from Unilateral–Given Friends OG	.00	.03	.06	.16***	.13***
7. Child-Level Sum Provisions Received from Unilateral–Received Friends SG	.00	−.06	.02	.02	−.31***	−.07*
8. Child-Level Sum Provisions Received from Unilateral–Received Friends OG	.00	−.12***	.09**	.17***	−.07*	−.04	.12***
9. Child-Level Sum Provisions Received from Non-Friend Classmates SG	.00	−.12***	−.20**	−.01	−.25***	−.09**	.12***	.08*
10. Child-Level Sum Provisions Received from Non-Friend Classmates OG	.00	−.17***	.04	.02	−.07*	−.10**	.15***	.10***	−.06
11. Child-Level Loneliness	−.13***	−.06	−.15**	−.01	.01	.04	−.07*	−.02	.01	−.03

Note. SG = same-gender, OG = other-gender.

^*p < .05

^**p < .01

^***p < .001.

First, the intercept-only model was estimated to decompose between- and within-classroom variability in loneliness. The ICC was .029, indicating that 2.9% of the total variance in loneliness was between classrooms, and 97.1% of the variance in loneliness was within classrooms (i.e., between children).

Controlling for grade and gender, sum provisions received from reciprocal friends (same- and other-gender) were entered in the first step of the model. Sum provisions from same-gender, but not other-gender, reciprocal friends was a significant predictor of loneliness. The χ² difference test with four degrees of freedom indicated that the first step of the model was a significant improvement in fit from the intercept-only model (p < .001). The first step of the model explained 3.1% of the 97.1% variance in loneliness that was at the child level, and 53.0% of the 2.9% variance in loneliness that was at the classroom level (Table 8).

In the second step of the model, sum provisions from same-gender, but not other-gender, unilateral–received friends was a significant predictor of loneliness, over and above the contribution of sum provisions from reciprocal friends. Sum provisions from unilateral–given same- and other-gender friends were not significant. The χ² difference test with four degrees of freedom indicated improvement in model fit (p = .037). This step explained a small additional 1.1% of child-level variance in loneliness and did not explain additional classroom-level variance in loneliness (Table 8).

The third step of the model added in sum provisions from same- and other-gender non-friend classmates of which neither were significant predictors of loneliness. The χ² difference test with two degrees of freedom did not indicate significant improvement in model fit (Table 8).

To aid interpretation of these findings, we calculated the change in loneliness that would be associated with a 1 SD increase in a given significant predictor (see Rumberger & Palardy, 2004, for a discussion of this approach). A 1 SD increase in provisions received from same-gender reciprocal friendships is associated with a 1.49 unit decrease in loneliness (measured on a 12–60 scale). A 1 SD increase in provisions received from same-gender unilateral–received friendships is associated with a 0.67 unit decrease in loneliness.

Goal 3: Child- and Classroom-Level Predictors of Loneliness

Next, we examined whether classroom-level provision richness helped explain variation in children’s feelings of loneliness over and above the provisions that children individually received (Goal 3a) and whether classroom-level provision richness moderated the child-level effect of provisions on loneliness (Goal 3b). Parameter estimates, fit statistics, and variance components for all models are presented in Table 10, including the intercept-only model, which as noted above indicated that the ICC for loneliness was .029. Correlations among variables included in these analyses are in Table 11.

Table 10.

Parameter Estimates from Multilevel Regressions Predicting Children’s Loneliness from Child- and Classroom-Level Provisions Received

	Intercept-Only Model	Child-Level Provisions Received	Classroom-Level Provisions Received	Child- x Classroom-Level Interaction
	Estimate (se)	Estimate (se)	Estimate (se)	Estimate (se)
Intercept	17.36 (0.33)***	19.84 (0.60)***	19.75 (0.57)***	19.77 (0.57)***
Child Gender (Child Level)		−1.70 (0.51)**	−1.70 (0.51)**	−1.73 (0.51)***
Grade (Classroom Level)		−1.32 (0.37)**	−1.24 (0.35)**	−1.23 (0.35)**
Child-Level Provisions Received SG		−11.91 (1.99)***	−11.90 (1.99)***	−12.60 (2.02)***
Child-Level Provisions Received OG		−0.37 (5.16)	−0.41 (5.16)	−0.30 (5.15)
Classroom-Level Provisions Received SG			−12.58 (6.01)*	−12.43 (6.02)*
Classroom-Level Provisions Received OG			15.35 (8.31)	15.62 (8.32)
Classroom- x Child-Level Provisions Received SG				71.65 (35.28)*
τ00	1.83 (1.00)*	0.82 (0.73)	0.50 (0.65)	0.52 (0.65)
σ 2	62.20 (2.86)***	59.77 (2.75)***	59.74 (2.74)***	59.48 (2.73)***
AIC	6905.10	6863.90	6863.30	6861.20
BIC	6910.10	6875.40	6878.00	6877.60
Deviance (−2 log likelihood)	6899.10	6849.90	6845.30	6841.20
Δχ2		49.20***	4.60	4.10*

Note. Models were estimated in SAS 9.3 using PROC MIXED with maximum likelihood estimation (substantive conclusions are identical using restricted maximum likelihood estimation).

Child gender is coded 0 = girls, 1 = boys; grade is coded 0 = 3^rd grade, 1 = 4^th grade, 2 = 5^th grade; SG = same-gender, OG = other-gender; se = standard error. Child-level provisions scores are mean-centered within classrooms, classroom-level provisions scores are mean-centered at the grand mean. The intercept was modeled as a random effect, all others are fixed effects; τ₀₀ = estimate of variance at Level 2 (classroom level); σ² = estimate of variance at Level 1 (child level). AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion.

^*p < .05

^**p < .01

^***p < .001.

Table 11.

Correlation Among Variables Included in Multilevel Regression Predicting Children’s Loneliness from Child- and Classroom-Level Provisions Received (Goal 3)

	1.	2.	3.	4.	5.	6.
1. Grade
2. Gender	.01
3. Child-Level Provisions Received SG	−.01	−.26***
4. Child-Level Provisions Received OG	−.01	−.13***	.12***
5. Classroom-Level Provisions Received SG)	.10***	−.02	−.02	−.02
6. Classroom-Level Provisions Received OG	.04	−.04	−.02	−.02	.64***
7. Child-Level Loneliness	−.13***	−.06	−.16***	−.01	−.05	.03

Note. SG = same-gender, OG = other-gender.

^***p < .001.

Child-level provisions received proportion scores from same- and other-gender peers (centered within classroom) were entered as predictors of loneliness, controlling for child gender and grade. Consistent with the analyses by relationship source presented above, provisions received from same-gender peers, but not other-gender peers, were associated with loneliness, such that children who received relatively more provisions from same-gender classmates were less lonely. The χ² difference test with four degrees of freedom indicated a significant improvement in model fit from the intercept-only model (p < .001). The first step of the model explained 3.9% of the 97.1% variance in loneliness that was at the child level, and 55.2% of the 2.9% variance in loneliness that was at the classroom level (Table 10).

In the next step of the model, we examined whether classroom provision richness would predict children’s feelings of loneliness over and above the provisions they received from classroom peers. To address this question, classroom-level averages of provisions received from same- and other-gender peers (centered at the sample mean) were entered as predictors of loneliness. Controlling for child-level provisions received, same-gender classroom provision richness, but not other-gender provision richness, was a significant predictor of loneliness. However, the χ² difference test with two degrees of freedom indicated no significant improvement in model fit, suggesting that the main effects of same- and other-gender provision richness were not large enough to add significantly to the prediction of loneliness.

Nevertheless, the main effect of same-gender classroom provision richness should be interpreted in the context of the interaction effect tested in the next step. In the next step of the model, we examined whether classroom provision richness moderated the effect of child-level provisions received on loneliness. That is, does being in a relatively provision-rich classroom buffer or exacerbate the effects of receiving relatively few provisions on feelings of loneliness? To address this question, we estimated the cross-level interaction between child-level provisions received and classroom provision richness. Because children received relatively few provisions from other-gender peers, and provisions received from other-gender peers were not a significant predictor of loneliness at either level, we estimated the cross-level interaction for provisions received from same-gender peers only. The interaction between child- and classroom-level provisions received from same-gender peers was statistically significant, as was the χ² difference test with one degree of freedom (p = .043), indicating that the effect on loneliness of provisions received from same-gender peers at the child level was moderated by the provision richness of the classroom. This step explained an additional .44% of the variance in child-level loneliness but did not explain additional classroom-level variance in loneliness.

To characterize the moderation effect, we estimated simple slopes for the effect of child-level provisions received in classrooms that were low (20^th percentile; M = .30), average (50^th percentile; M = .34), and high (80^th percentile; M = .40) in same-gender provisions received using Preacher et al.’s (2006) interaction utility for multilevel regression. Although the effect of child-level provisions received on loneliness was significantly different from zero at all levels of provision richness, the association was less pronounced in more provision-rich classrooms. That is, as shown in Figure 1(a), child-level provisions received were less strongly related to loneliness in classrooms that were more provision-rich.

Figure 1. Graphical Representation of the Interaction Between Child- and Classroom-Level Provisions Received from Same-Gender Classmates.

Note. In (a), effects of provisions received from same-gender (SG) classmates are plotted for classrooms that are low, average, and high in provisions received from same-gender classmates. For low-provisions classrooms, simple intercept = 20.39 (se = 0.63), p < .001, simple slope = −16.16 (se = 2.89), p < .001; for average-provisions classrooms, simple intercept = 19.81 (se = 0.57), p < .001, simple slope = −12.86 (se = 2.05), p < .001; for high-provisions classrooms, simple intercept = 19.11 (se = 0.68), p < .001, simple slope = −8.83 (se = 2.50), p < .001. In (b), effects of classroom-level provisions received from same-gender classmates are plotted for children who are low, average, and high in provisions received from same-gender classmates. For low-provisions children, simple intercept = 21.19 (se = 0.64), p < .001, simple slope = −20.53 (se = 7.18), p = .004; for average-provisions children, simple intercept = 19.82 (se = 0.57), p < .001, simple slope = −12.71 (se = 6.02), p = .03; for high-provisions children, simple intercept = 18.37 (se = 0.59), p < .001, simple slope = −4.46 (se = 7.23), p = .537.

To directly address the “loneliness buffering” versus “loneliness exacerbating” hypotheses, we probed the interaction in the other direction, estimating simple slopes for the effect of classroom-level provision richness at low (20^th percentile, M = .22), average (50^th percentile, M = .34), and high (80^th percentile, M = .47) levels of individual same-gender provisions received. The effect of classroom-level provision richness was significant for low or average individual provisions received, but not for high individual provisions received. As shown in Figure 1(b), these findings indicate that classroom provision richness buffered against loneliness for children who individually received fewer provisions, supporting the buffering rather than exacerbating hypothesis. In interpreting these findings, it is important to keep in mind that, because child-level provisions scores are centered within classroom, children should be considered as high or low in provisions received relative to others in their class, as opposed to others within the sample as a whole (see Appendix Table A4 for descriptive information about provisions received separately for each class).

To aid interpretation, we also calculated the change in loneliness that would be associated with a one standard deviation change in a predictor. A 1 SD change in child-level provisions received from same-gender peers resulted in a 1.78 unit decrease in loneliness. A 1 SD change in same-gender classroom-level provision richness resulted in a 0.73 unit decrease in loneliness. These effects are small but provide evidence that classroom-level provision richness is associated with children’s experience in the classroom, and the interaction effects suggest this association is especially true for children who are lower in child-level provisions received.

Discussion

The present study is the first to assess the social provisions children receive from each of their classmates, and to examine the associations between individual provisions received, classroom provision richness, and children’s feelings of loneliness in school. We found that children received provisions from classmates with whom they shared varying types of relationships and not just from reciprocal friends, but that only receipt of provisions from reciprocal friends and unilateral–received friends were significantly associated with lower levels of loneliness. When looking at the classroom level, classroom provision richness moderated the association between social provisions receipt and feelings of loneliness, with this effect being strongest for children at lower levels of individual provisions received. Collectively, the findings suggest that capturing a fuller picture of the social provision landscape of the classroom is important for understanding children’s social-emotional well-being in elementary school. In the sections below, we discuss key findings, their implications for research and application, as well as future directions.

Examining the Classroom Social Provisions Landscape (Goal 1)

Gender Match and Relationship Type Differences (Goals 1a and 1b)

This study provided the opportunity to examine the exchange of provisions among same-gender and other-gender classmates in elementary school. As expected, receipt of social provisions within the classroom context was highly gender segregated, with children receiving only 7% of their social provisions from other-gender classmates. These data highlight how much children may be missing out by living in a social world in which they primarily interact with same-gender peers and thereby receive relatively few social provisions from other-gender peers. The gender divide in peer interaction emerges early for both human children (Maccoby, 1998) and non-human primates (Rosenblum et al., 1975), but it might be influenced by contextual factors and the degree of shared interests. To illustrate this possibility, Mulvey et al. (2020) found that having boys and girls participate together in a motor competence intervention increased the inclination of children, especially girls, to view other-gender peers as part of their friendship group.

Additionally, when we examined the (albeit limited) provisions children received from other-gender peers in terms of the specific types of provisions children received, we found that help and validation were more common than opportunities for self-disclosure or companionship. These findings suggest that facilitating further opportunities for help and validation with other-gender classmates in the classroom might be a natural first step for increasing positive interactions and feelings among mixed-gender groups of children given that these interactions are already happening, at least to some degree.

We also found that children received more than half of their social provisions from classmates who were not reciprocal friends. This finding suggests that researchers should look beyond reciprocated friendships when assessing a child’s receipt of social provisions. Furthermore, examining provision exchange among all classmates allows for a more complete assessment of classroom provision richness, which we found had implications for children’s classroom experience.

Classroom-Level Correlates of Provision Richness (Goal 1c)

The present study also provides evidence about the sociometric and behavioral correlates of provision-rich classrooms. Provision-rich classrooms were also classrooms in which children, on average, were more highly accepted by their peers. However, contrary to our expectation, provision rich classrooms were not characterized by higher levels of reciprocated friendship ties. This pair of findings are consistent with previous theoretical and empirical work distinguishing peer acceptance from the number of friends children have (Bukowski et al., 1993; Gest et al., 2001; Kingery et al., 2011; Parker & Asher, 1993; see Asher & Weeks, 2018, for a recent discussion). From a “social tasks” perspective on friendship, the processes involved in developing and maintaining a close friendship go beyond the processes involved in becoming well-liked by peers (for analyses and discussion of the social tasks of friendship see Asher et al., 1996, Flannery & Smith, 2016, MacEvoy & Asher, 2012, and Rose & Asher, 2017). Although being in a provision-rich classroom may facilitate higher levels of peer liking and thereby provide a foundation for friendship formation, classroom provision richness may be insufficient by itself to result in the formation of more friendships within a classroom.

As for the social behavioral correlates of provision-rich classrooms, classrooms high in provision richness were also higher in average levels of prosocial behavior. This finding is not surprising given the conceptual overlap between these constructs, as discussed in the introduction, but this association between provision richnesss and prosocial behavior remained even when help was removed from the classroom provision richness measure. This finding, and the correlation with peer acceptance, also provide preliminary evidence for concurrent validity of the new peer provision nomination measure.

Contrary to our expectations, provision-rich classrooms were not lower in average levels of aggressive or withdrawn behavior. The lack of correlation concurrently does not rule out the possibility that over time a provision-rich classroom might also develop lower levels of aggression or withdrawn behavior. Such longitudinal work is an important future direction. Additionally, despite having a large sample of individual children and the number of classrooms being comparable to other studies where classrooms as well as individual children are the units of analysis (e.g., Chung-Hall & Chen, 2009; Kilday & Ryan, 2019; Kim & Cappella, 2016; Morrow et al., 2019), more statistical power at the classroom level is needed to determine whether modest associations do exist between classroom provision richness and the less frequent classroom behaviors of aggression and withdrawal, or to confirm that classroom provision richness is indeed not associated these types of behaviors.

Social Provisions Beyond Reciprocal Friends and Associations with Loneliness (Goal 2)

Social provisions from reciprocal friends were significantly associated with feelings of loneliness, which is consistent with prior research linking having a single high-quality very best friendship with lower feelings of loneliness (e.g., Parker & Asher, 1993). A key contribution of the present study was examining receipt of provisions from all reciprocal friends in the classroom and their association with loneliness. We also found that provisions received from unilateral–received friends were associated with lower loneliness. Although this association was small, it suggests that relationships beyond reciprocal friendships can play an important role in meeting a child’s social needs.

The Effect of Classroom Provision Richness (Goal 3)

Regarding Goal 3a, our findings did not support the hypothesis that there was a classroom effect of provision richness on children’s loneliness beyond the effect of children’s individual receipt of provisions, however we did find a moderation effect of classroom provision richness (Goal 3b). With regard to Goal 3b, we offered competing hypotheses about how the provision richness of a classroom might be relevant for the association between individual provisions received and loneliness. Findings indicated that provision-rich classrooms buffered the effects on loneliness for children individually receiving fewer provisions, rather than supporting the alternative prediction that children who receive fewer provisions than their peers would feel more lonely, perhaps due to social comparison processes. A provision-rich environment may convey a sense of warmth and the potential for future positive social interactions to children who are currently receiving fewer provisions from classmates. It is interesting that classroom provision richness did not have the same degree of relevance for children who personally received many provisions, possibly because their provision needs are sufficiently met by their interactions with classmates.

Building on the interest of researchers and educators in the effects of classroom climate on social development (e.g., Hamre & Pianta, 2010; Wentzel & Ramani, 2016), findings from this study provide evidence that classroom provision richness can be reliably measured and does relate to loneliness, especially for children who themselves receive fewer provisions from classmates. Even though the overall effect size was small, the effect for the students who were receiving fewer provisions from classmates and were more lonely may be consequential. The experience of loneliness is multiply determined and has potentially serious consequences, so understanding contributing factors—and identifying potential buffering effects—even with small effect sizes, can be meaningful for children and for getting a fuller picture of the factors that are associated with loneliness. Indeed, Götz and colleagues (2022) recently highlighted both “the empirical relevance and practical significance” of small effect sizes for building a cumulative and replicable body of evidence in psychological science (p. 3). Furthermore, effect sizes could well be larger if deliberate efforts were made to foster provision richness in schools. Our hope is that these findings will spark future research focused on how provision-rich classrooms can be fostered and how their protective effects against loneliness can be amplified.

Future Directions and Implications

Examining Provision-Specific Associations with Loneliness

Building on previous research and theory on developmental changes in children’s conceptions of peer relationships (e.g., Bigelow & LaGaipa, 1975; Clark & Bittle, 1992; Furman & Bierman, 1984; Parker & Gottman, 1989), Parkhurst and Hopmeyer (1999) theorized that certain provisions might be especially relevant for loneliness at different points in development, with companionship and help being important in childhood, and more developmentally “advanced” provisions such as validation and self-disclosure gaining importance in adolescence. Due to the multilevel nature of our data and our focus on cross-level interactions, in the present study we were unable to fully test Parkhurst and Hopmeyer’s (1999) hypothesis about the relative importance of some provisions over others in middle childhood, but we conducted exploratory multilevel analyses separately for each of the four types of provisions assessed in this study (see Supplementary Materials page 2, Tables S2–S6, and Figures S1 and S2). We found that each child-level provision was significantly associated with loneliness to varying degrees, but significant classroom-level moderation effects emerged only for companionship and help. We include these supplemental findings as a reference point for future research on the relative influence of specific provisions for loneliness in childhood.

Provision Richness in the Broader Classroom Context

The present findings also extend research (e.g., Boor-Klip et al., 2017; Downer et al., 2007; Kim & Cappella, 2016; Rohlf et al., 2016; Torrente et al., 2014) on classroom-level moderators of children’s experiences in school by characterizing, for the first time, the classroom social environment in terms of provision richness. Likewise, this study appears to be the first to examine the role of the classroom context in elementary-school children’s loneliness in school. Future research is needed to examine how the provision richness of a classroom is associated with other aspects of children’s emotional lives, such as the level of disappointment and hurt feelings or even resentment and anger that children experience when their classroom lives are under-provisioned. Research is also needed on how the provision richness of a classroom relates to children’s academic engagement. As previously mentioned, diverse features of the classroom social environment have been linked with academic engagement and competency outcomes (e.g., Cappella et al., 2013; Chung-Hall & Chen, 2009; Kilday & Ryan, 2019; Kim & Cappella, 2016; Torrente et al., 2014). In one study, classroom-level prosocial behavior and academically oriented behaviors moderated connections between children’s prosocial behavior and peer acceptance, and between their academic behavior and peer acceptance (Torrente et al., 2014). Classroom provision richness might also have direct or moderating effects on academic outcomes. Conceivably, a student in a classroom where children engage in companionship, help, validation, and self-disclosure with many peers might have a higher sense of belonging, a less negative emotional experience, and be more able to engage in learning than a student in a classroom where provisions are more selectively exchanged. This possibility aligns with theoretical and empirical work suggesting that students’ motivation and engagement in learning is at least in part due to involvement in caring social relationships in the classroom (see Wentzel, 2022, for a review). Additionally, a provision-rich classroom may also promote a child’s cognitive development by establishing a more supportive interaction environment for learning and fostering positive peer connections with time. Microgenetic research using high-density measurement of collaborative reasoning interactions in elementary classrooms (Lin et al., 2015) found that support directed toward the child, or observed occurring between peers, was associated with growth in relational thinking (i.e., being able to conceive of and manipulate relations between concepts).

Fostering Provision Richness

The implications of classroom provision richness for loneliness and potentially other outcomes such as achievement suggest that an important future direction is to examine how educators can enhance the provision richness of their classrooms. Previous work has highlighted the role that teachers can play in shaping the peer environment (e.g., Farmer et al., 2011; Gest & Rodkin, 2011; Hanish et al., 2016; Hendrickx et al., 2017), and teacher cultivation of a classroom environment characterized by warmth, social opportunity, and open exchange is likely to be a key factor in facilitating classroom provision richness. Teachers who engage in specific practices that promote children helping, supporting, and providing companionship to peers more widely, not just toward friends, are likely to have more provision-rich classrooms. Additionally, the individual teacher–student relationship itself may also be influential in promoting provision richness. In adolescence, students’ perceptions of teacher support were found to have direct effects on students’ perceptions of supportive peer environments (Morin, 2020), and such effects likely extend to younger students as well. Teacher emotional support has been longitudinally linked with children’s prosocial and cooperative behavior in class (Luckner & Pianta, 2011), and individual- and classroom-level student–teacher relationship quality has been associated with children’s prosocial and aggressive behavior and well-being (Buyse et al., 2009; Rudasill et al., 2013). Whether such teacher–student relationship features similarly foster provision-rich classrooms through modeling behaviors, establishing classroom norms, or other processes is an important question for future research.

Methodology: Advances and Extensions

The present study provides a solid foundation for these future directions as it capitalizes on several methodological strengths. We avoided potential confounds between the assessment of social provisions and the assessment of loneliness by (a) using a highly focused measure of loneliness that asks only about feelings of loneliness without asking about hypothesized causes of loneliness (such as the receipt of provisions from social relationships), and (b) using a novel peer nomination method to examine the exchange of social provisions within the classroom and therefore avoiding shared method variance in the assessment of key constructs. Furthermore, the peer nomination social provisions measure provides reliable information about the provisions children receive from all classmates, not just a single close friend.

Another key methodological strength was the high participation rate of 98% of children from 38 classrooms at two schools. This high rate of participation enabled a comprehensive and valid assessment of the social provisions children received in the classroom, which in turn allowed for the examination of classroom provision richness and its potential associations with children’s loneliness.

Limitations

The community and schools in which this research was conducted were fairly racially homogenous, so it will be important to examine the degree to which these findings replicate in more diverse school contexts. Future research with even more classrooms would allow for the examination of potential moderators such as the racial homogeneity vs. heterogeneity of the classroom. As research on classroom ethnic diversity, other-group peer relationships, and intergroup attitudes in school (e.g., Davies et al., 2011; Graham et al., 2014; Juvonen et al., 2018; Van Ryzin et al., 2020) continues to grow, attention to the broader exchange of social provisions in racially heterogeneous contexts will likely be informative.

Additionally, the present study categorized gender as a binary variable, which overlooks important gender variation. Future research should attend to the potential role of gender diversity in the classroom. Gender identity might be related to receipt of and sources of social provisions, which in turn might have downstream consequences for students’ feelings of loneliness.

Conclusion

Elementary school children in most parts of the world spend much of their day in self-contained classroom environments. The concept of classroom provision richness provides a way to characterize a distinct facet of the classroom peer environment by building on previous research and theory on social provisions within friendships to learn about the wider peer context of classroom life and its association with children’s feelings of loneliness. This research will hopefully stimulate others to consider the role that social provisions might play at multiple levels of analysis in explaining not only feelings of loneliness, but also other important indices of children’s social, emotional, and academic functioning in school.

Appendix

Table A1.

Grade Differences in Loneliness and Social Provisions

	M (SD)
	Third	Fourth	Fifth	df	F	Partial η2
Loneliness	19.17 (9.50)	17.33a (8.21)	16.47a (6.84)	(2, 981)	8.56***	.017
Social Provisions Overall SG	.34 (.14)	.34 (.14)	.35 (.14)	(2, 992)	.96	.002
Social Provisions Overall OG	.08a (.06)	.06 (.06)	.07a (.06)	(2, 992)	9.66***	.019
Companionship SG	.43 (.20)	.41 (.17)	.41 (.17)	(2, 992)	1.13	.002
Help SG	.34a (.15)	.37ab (.17)	.40b (.16)	(2, 992)	8.24***	.016
Validation SG	.40 (.18)	.37 (.18)	.38 (.18)	(2, 992)	1.87	.004
Self-disclosure SG	.20a (.14)	.20a (.13)	.22 (.15)	(2, 992)	4.73**	.009
Companionship OG	.03a (.06)	.02ab (.06)	.02b (.04)	(2, 992)	3.85*	.008
Help OG	.12a (.10)	.09 (.09)	.13a (.11)	(2, 992)	15.69***	.031
Validation OG	.12a (.09)	.09 (.10)	.11a (.10)	(2, 992)	6.07**	.012
Self-disclosure OG	.03ab (.05)	.02a (.05)	.03b (.06)	(2, 992)	7.24***	.014

Note. Social provision scores are provisions received proportion scores. Ns for loneliness were lower than for the other variables due to missing data: third grade n = 212, fourth grade n = 335, fifth grade n = 440. For all other variables in the table, Ns were the full sample: third grade n = 216, fourth grade n = 341, fifth grade n = 441. In cases of significant grade differences, grades that share superscripts are not significantly different from each other at the p < .05 level.

SG = same-gender, OG = other-gender.

^*p < .05,

^**p < .01,

^***p < 001.

Table A2.

Gender Differences in Loneliness and Social Provisions

	M (SD)
	Girls	Boys	df	F	Partial η2
Loneliness	17.85 (7.84)	16.89 (8.12)	(1, 981)	4.79*	.005
Social Provisions Overall SG	.38 (.15)	.31 (.13)	(1, 992)	61.83***	.059
Social Provisions Overall OG	.08 (.07)	.06 (.06)	(1, 992)	18.56***	.018
Companionship SG	.41 (.18)	.42 (.18)	(1, 992)	0.04	.000
Help SG	.41 (.17)	.34 (.15)	(1, 992)	49.52***	.048
Validation SG	.44 (.18)	.33 (.16)	(1, 992)	86.91***	.081
Self-disclosure SG	.26 (.15)	.16 (.12)	(1, 992)	131.83***	.117
Companionship OG	.03 (.06)	.02 (.05)	(1, 992)	9.41**	.009
Help OG	.13 (.11)	.10 (.10)	(1, 992)	35.34***	.034
Validation OG	.11 (.10)	.10 (.10)	(1, 992)	1.17	.001
Self-disclosure OG	.04 (.06)	.02 (.05)	(1, 992)	17.38***	.017

Note. Social provision scores are provisions received proportion scores. Ns for loneliness were lower than for the other variables due to missing data: girls n = 461, boys n = 526. For all other variables in the table, Ns were the full sample: girls n = 468, boys n = 530.

SG = same-gender, OG = other-gender.

^*p < .05,

^**p < .01,

^***p < 001.

Table A3.

Grade by Gender Differences in Loneliness and Social Provisions Variables

	M (SD)
	Girls			Boys
	Third	Fourth	Fifth	Third	Fourth	Fifth	Gender x Grade df	Gender x Grade F	Partial η2
Loneliness	20.49 (9.81)	17.44 (7.59)	16.89 (6.61)	18.00 (9.10)	17.23 (8.73)	16.11 (7.03)	(2, 981)	1.39	.003
Social Provisions Overall SG	.38 (.16)	.38 (.15)	.38 (.14)	.31 (.11)	.30 (.13)	.33 (.13)	(2, 992)	1.63	.003
Social Provisions Overall OG	.10 (.06)	.06 (.07)	.08 (.07)	.06 (.05)	.06 (.05)	.07 (.06)	(2, 992)	5.77**	.011
Companionship SG	.44 (.21)	.41 (.17)	.40 (.17)	.42 (.20)	.41 (.18)	.42 (.18)	(2, 992)	1.14	.002
Help SG	.37 (.18)	.44 (.17)	.42 (.17)	.32 (.13)	.31 (.15)	.34 (.15)	(2, 992)	8.21***	.016
Validation SG	.46 (.18)	.43 (.19)	.43 (.18)	.35 (.16)	.32 (.16)	.34 (.16)	(2, 992)	0.62	.001
Self-disclosure SG	.26 (.17)	.25 (.15)	.27 (.15)	.14 (.09)	.15 (.10)	.18 (.13)	(2, 992)	0.45	.001
Companionship OG	.04 (.06)	.03 (.06)	.03 (.04)	.03 (.05)	.02 (.05)	.02 (.04)	(2, 992)	1.19	.002
Help OG	.17 (.10)	.10 (.10)	.15 (.11)	.08 (.08)	.09 (.08)	.12 (.11)	(2, 992)	12.12***	.024
Validation OG	.13 (.09)	.09 (.09)	.12 (.10)	.11 (.09)	.10 (.10)	.11 (.10)	(2, 992)	2.30	.005
Self-disclosure OG	.04 (.05)	.02 (.05)	.05 (.07)	.02 (.05)	.02 (.04)	.02 (.05)	(2, 992)	4.30*	.009

Note. Social provision scores are provisions received proportion scores. Ns for loneliness were lower than for the other variables due to missing data: third grade girls n = 100, fourth grade girls n = 156, fifth grade girls n = 205; third grade boys n = 112, fourth grade boys n = 179, fifth grade boys n = 235. For all other variables in the table, Ns were the full sample.

SG = same-gender, OG = other-gender.

^*p < .05,

^**p < .01,

^***p < 001.

Table A4.

Descriptive Statistics for Loneliness and Same- and Other-Gender Provisions Received by Class

			Class-level Provision Richness		Child-level SG Provisions Received			Child-level OG Provisions Received			Loneliness
Class #	Class Size	Grade	SG M	OG M	Range	M	SD	Range	M	SD	Range	M	SD
1	23	5	0.23	0.04	0.06–0.38	0.23	0.09	0–0.13	0.04	0.04	12–35	15.87	5.56
2	27	5	0.25	0.05	0.1–0.43	0.25	0.10	0–0.1	0.05	0.03	12–45	18.89	8.71
3	27	4	0.25	0.02	0.07–0.46	0.25	0.11	0–0.07	0.02	0.02	12–60	18.96	10.81
4	26	4	0.27	0.02	0.09–0.54	0.27	0.12	0–0.07	0.02	0.02	12–42	19.08	7.89
5	29	4	0.27	0.01	0–0.5	0.27	0.12	0–0.17	0.01	0.03	12–49	17.79	8.93
6	29	4	0.28	0.03	0–0.64	0.28	0.12	0–0.32	0.03	0.07	12–60	18.15	11.10
7	23	3	0.29	0.04	0.05–0.45	0.29	0.10	0.01–0.11	0.04	0.03	12–49	21.13	10.78
8	23	3	0.30	0.06	0.1–0.48	0.30	0.11	0.02–0.16	0.06	0.04	12–43	19.83	8.84
9	25	4	0.30	0.05	0.05–0.48	0.30	0.10	0–0.13	0.05	0.04	12–32	17.36	6.52
Low provision rich classroom 20th percentile mark = .30
10	28	4	0.31	0.02	0.06–0.54	0.31	0.11	0.01–0.07	0.02	0.01	12–43	16.81	6.91
11	24	3	0.31	0.06	0.07–0.55	0.31	0.12	0–0.22	0.06	0.05	12–56	22.67	12.39
12	30	5	0.31	0.15	0.13–0.56	0.31	0.10	0.04–0.37	0.15	0.09	12–42	15.93	7.37
13	26	3	0.32	0.04	0.1–0.78	0.32	0.15	0–0.16	0.04	0.03	12–37	17.08	6.90
14	27	5	0.32	0.03	0.05–0.52	0.32	0.13	0–0.08	0.03	0.02	12–29	15.96	5.26
15	27	5	0.32	0.04	0.01–0.64	0.32	0.16	0–0.15	0.04	0.04	12–38	15.48	6.32
16	25	4	0.33	0.09	0.15–0.56	0.33	0.12	0.04–0.21	0.09	0.05	12–43	18.80	8.18
17	30	5	0.33	0.08	0.08–0.56	0.33	0.13	0.02–0.18	0.08	0.04	12–34	17.00	6.63
18	28	5	0.34	0.04	0.05–0.6	0.34	0.15	0–0.13	0.04	0.03	12–29	14.81	4.24
19	30	5	0.34	0.05	0.17–0.54	0.34	0.10	0–0.18	0.05	0.04	12–28	15.27	3.93
Average provision rich classroom 50th percentile mark = .34
20	23	3	0.35	0.04	0.1–0.63	0.35	0.15	0.03–0.1	0.04	0.02	12–47	18.96	10.84
21	26	4	0.36	0.05	0.12–0.56	0.36	0.11	0.01–0.21	0.05	0.04	12–27	16.35	4.66
22	24	3	0.36	0.15	0.22–0.58	0.36	0.10	0.06–0.27	0.15	0.06	12–46	17.74	7.77
23	24	4	0.36	0.09	0.17–0.61	0.36	0.14	0.03–0.15	0.09	0.04	12–38	14.33	5.73
24	29	5	0.36	0.07	0.16–0.52	0.36	0.10	0.01–0.15	0.07	0.04	12–60	19.83	11.50
25	26	3	0.37	0.06	0.11–0.81	0.37	0.20	0.02–0.09	0.06	0.02	12–40	17.62	8.02
26	27	4	0.38	0.05	0.14–0.68	0.38	0.12	0–0.14	0.05	0.03	12–25	15.11	4.22
27	26	5	0.39	0.11	0.15–0.69	0.39	0.14	0.03–0.37	0.11	0.09	12–37	14.85	5.98
28	25	5	0.39	0.05	0.12–0.63	0.39	0.15	0.01–0.15	0.05	0.03	12–21	14.48	3.44
29	22	5	0.39	0.06	0.08–0.68	0.39	0.19	0.01–0.13	0.06	0.04	12–45	15.86	7.73
30	28	5	0.40	0.07	0.07–0.68	0.40	0.15	0–0.19	0.07	0.06	12–28	17.39	5.69
31	23	3	0.40	0.14	0.22–0.66	0.40	0.14	0.03–0.35	0.14	0.09	12–57	20.70	12.32
32	30	5	0.40	0.08	0.19–0.58	0.40	0.12	0.03–0.19	0.08	0.04	12–35	18.60	6.46
33	24	3	0.40	0.09	0.1–0.6	0.40	0.14	0–0.31	0.09	0.07	12–29	17.00	4.80
High provision rich classroom 80th percentile mark = .40
34	25	4	0.42	0.07	0.12–0.78	0.42	0.19	0.03–0.14	0.07	0.03	12–22	14.18	2.99
35	25	4	0.43	0.07	0.06–0.71	0.43	0.20	0–0.39	0.07	0.08	12–52	16.28	10.02
36	29	5	0.45	0.14	0.18–0.69	0.45	0.11	0.04–0.29	0.14	0.08	12–33	15.10	5.63
37	30	5	0.45	0.14	0.14–0.68	0.39	0.11	0.08–0.36	0.13	0.05	12–45	17.43	8.78
38	25	4	0.48	0.19	0.28–0.67	0.48	0.08	0.11–0.37	0.19	0.06	12–49	21.48	11.10

Note. Classes are sorted from lowest to highest proportion of class-level same-gender provision richness.

SG = same gender, OG other gender. Grade is coded as 3 = third grade, 4 = fourth grade, 5 = fifth grade.