1. Introduction

During adolescence, building relationships with peers and developing social competencies are of crucial importance for well-being. However, adolescents with chronic pain (ACP) often experience impairments in their social functioning.25,34 They report more frequent school absence and greater social isolation, be it through social exclusion by others or self-withdrawal.51 This isolation may install a sense of loneliness, often associated with a perceived lack of understanding from peers who do not know what it is like to experience chronic pain.24,26 However, little is known about the consequences of reduced social functioning of ACP on their relationships with peers. Indeed, social networks of ACP are complex and consist of multiple peer dyads in which each member has their own perspective on the same relationship.

A social network approach provides a valuable framework to assess both self-perceptions and peer- perceptions of dyadic relationships of adolescents within their social network.25,50 Social network analysis (SNA) adds an advantage over traditional respondent-centered approaches because it allows for the identification of structural characteristics of peer relationships within specific environments (eg, a class or school).50 Earlier network research in other contexts has identified 3 characteristics, namely popularity, homophily, and relationship quality, which may be crucial in understanding the social networks of ACP.8,22

First, peer relationships involve an individual's popularity, referring to how often one is nominated as a friend by peers. Because the perception of ACP their own popularity may be biased, for instance, by negative attribution styles, including a peer perspective is important.27 Studies examining popularity in ecologically valid contexts (eg, a school) are sparse. Second, individuals show a tendency to maintain relationships with others who resemble them regarding relevant characteristics, which is described as the homophily principle (similarity-attraction theory).11,42 A perceived lack of understanding from peers without pain may motivate ACP to interact with peers who also experience pain because ACP may feel better understood by them. Nonetheless, it is still unclear whether this hypothesis holds true. Finally, examining the quality of relationships is also crucial.14 A perceived positive friendship quality has been associated with reduced feelings of depression and loneliness.33,44 However, friendships may act as a source of stress, as well as support,12 which may also have an impact on the mental distress of adolescents. Research into these characteristics from a social network perspective is largely lacking.

Considering these knowledge gaps, this study used SNA to examine the structural social network characteristics in peer relationships of ACP and addressed 4 research questions: (RQ1) Are adolescents with higher pain grades (ie, pain frequency, intensity, and disability) less often nominated by others as being part of a peer group? (popularity); (RQ2) Do adolescents with similar pain grades name each other more often as being part of the same peer group? (homophily); (RQ3) Do dyads with an adolescent experiencing a higher pain grade report less positive (eg, support) and more negative (eg, conflict) friendship qualities? (relationship quality); and (RQ4) Does a perceived positive or negative friendship quality moderate the relationship between pain and emotional distress? (relationship quality and mental distress). Because sex is important for individual consequences of pain,8 we additionally explored the role of sex in RQ1-4.

2. Methods 2.1. Participants

This study used data from the first wave (in 2014) of a larger longitudinal study (ie, The Three Cities study) examining risks and buffers influencing well-being throughout adolescence.9 This study followed up adolescents from 19 public schools. The schools were a mix of urban and rural schools, situated in 1 midsize city (156,000 inhabitants) and 2 small towns (26,000 and 30,000 inhabitants) in middle Sweden. All schools were public schools, broadly representing Swedish demographics. Data collection took place between 2014 and 2018, whereby only adolescents from grades 7 (at age 13 years) and 8 (at age 14 years) were included in the first wave of 2014.

Adolescents, as well as their parents, provided informed consent to participate in the longitudinal study. Parents of 3336 students were informed about the study through a written letter, after which parents who declined participation could send a form in a prepaid envelope indicating that they did not want their child to participate (N = 122; 3.6%). Consequently, not returning the form was regarded as passive informed consent. Adolescents actively provided informed consent before completing the questionnaires, after first receiving oral and written information about the study in the classroom. Four hundred forty-seven students did not agree to participate or were absent on the day of data collection, resulting in a sample of 2767 (82.9%) adolescents. They completed the questionnaires in the classroom in the presence of a trained test leader. No teachers were present in the room. Adolescents were given 90 minutes to complete the survey. As a compensation, each class received 300 Swedish crowns. The regional Ethical Board of Uppsala gave ethical approval for the study (No. 2013/384).

2.2. Measures 2.2.1. Sociodemographic characteristics

Adolescents answered questions about their sex, age, class, school grade, immigrant background, divorcement of parents, socioeconomic status, and which of the 19 schools they attended. Immigrant status was dichotomized as being born outside of Sweden or having both parents born outside of Sweden, following the official Swedish definition of immigrant background. Socioeconomic status was assessed using the Family Affluence Scale, version 2, from World Health Organization's Health Behavior in School-aged Children survey,1 which includes 4 items by which adolescents assessed their family's affluence. The cutoff score for low socioeconomic status was set to 4.61, which is 1 SD below the mean M = 6.28 (SD = 1.67) of Family Affluence Scale, version 2, in the Swedish Health Behavior in School-aged Children survey (N = 23,088).1 These scores were already used in a previous study on the same data set.4

2.2.2. Pain characteristics

This study focused on musculoskeletal pain, which was assessed through the following question: “How often during the last 6 months have you suffered from pain in the back/neck/shoulders?” This item was derived from the Health Behavior in School-Aged Children Checklist32 and rated on a scale from 1 (=rarely or never) to 5 (=almost every day). Afterwards, they were asked how painful their back/neck/shoulder pain was on average, ranging from 1 (=not at all painful) to 10 (=very painful). Pain interference was assessed by means of 3 items in which they were asked whether the pain had impaired their functioning during the last 6 months (1) at school, (2) during leisure activities, and (3) during contact with friends. Items were rated on a scale from 1 to 3 (1 = no; 2 = yes, some; and 3 = yes, definitely). A sum score was calculated for pain interference. Cronbach α in this study was 0.74.

Adolescents were categorized according to 5 pain grades61,62: grade 0 (ie, pain frequency rarely or never), grade I (ie, pain at least every month, low pain intensity [scores 1-5], and no or low pain interference [scores 3-5]), grade II (ie, pain at least every month, high pain intensity [scores 6-10], and no or low pain interference [scores 3-5]), grade III (ie, pain at least every month, high pain intensity [scores 6-10], and medium impairment [scores 6-7]), and grade IV (ie, pain at least every month, high pain intensity [scores 6-10], and high impairment [scores 8-9]). These pain questions and grades were validated in a study among the same Swedish adolescents from seventh to ninth grade.64

2.2.3. Social network questionnaire

To assess social networks within each school, adolescents were asked to think about the group of peers with whom they spent time at school. Adolescents could name up to 3 peers. This name generator approach has been previously used in research examining friendship relationships.40 Afterwards, several questions were asked about the peers they had named. First, adolescents indicated the class of each peer they mentioned. Second, they answered 3 questions about how positive the friendship quality was: “Supports you if you have fought with your parents or teacher,” “Stands up for you when others talk behind your back,” and “Cares about your feelings.” Finally, they also indicated how negative the relationship was by means of 3 questions: “Argues a lot,” “Get angry at each other often,” and “Often gets irritated with each other.” These 6 items were rated on a scale from 1 (=do not agree at all) to 5 (=agree completely). The positive and negative friendship quality items were derived from the Friendship Quality Questionnaire of Parker and Asher.48 A sum score was calculated for positive (range: 3-15) and negative friendship (range: 3-15) qualities. Cronbach α for the positive and negative friendship quality scale was 0.88 and 0.67, respectively.

2.2.4. Emotional distress

Emotional distress was assessed by means of 2 constructs, namely stress and depressive symptoms. We opted for depression and stress because friendship quality may be protective against depression and the negative effects of stress-inducing events such as peer victimization.47 The degree to which situations in adolescent's lives during the last month were appraised as stressful was assessed by means of the Perceived Stress Scale (PSS).13,53 Because we examined friendships and friendship quality among adolescents with and without pain, a more general measure of distress, such as the PSS, would allow for a broader understanding of the associations amongst pain, friendship quality, and distress when compared with specific forms of distress such as anxiety. The PSS includes 14 items, which are rated on a scale from 0 (=never) to 4 (=very often). The PSS has been found to be a reliable measure of stress in adolescents.45 A total sum score (range: 0-56) was calculated. Cronbach α in this study was 0.98.

The Center for Epidemiologic Studies Depression scale (CES-D)63 was used to assess depressive symptoms. The CES-D includes 20 items, and adolescents rated these items on a 5-point scale from 0 (=not at all) to 4 (=a lot). This scaling was adapted to the original 4-point scale using a standard transformation.58 The CES-D has been validated in this sample of Swedish adolescents.46 A sum score (range: 0-60) was calculated. Cronbach α in this study was 0.94.

2.3. Analyses

Because of the nature of the data collection process (ie, questioning all people in a specific context about their relationships with each other), statistical methods such as linear regression, which assume independent error terms between observations, cannot be applied.21 First, because observations in social networks are interdependent (and not randomly sampled from the general population), can affect each other (eg, through reciprocity and transitivity), and are part of the same social system (in this study: schools) and, second, because the data are collected in multiple schools and thus have a multilevel or multigroup structure.

Taking into account that observations are interdependent, a Multiple Regression Quadratic Assignment Procedure (MRQAP)38 was used to test the RQs regarding the social networks of ACP (RQ1-3). Multiple Regression Quadratic Assignment Procedure analysis is a state-of-the-art SNA method used to investigate associations between characteristics of dyads (eg, the level of pain homophily) and a binary or continuous tie variable (eg, friendship nomination and level of perceived positive or negative friendship quality).15,37,38 Multiple Regression Quadratic Assignment Procedure provides a solution for the violation of the independence assumption by permuting random dependent network matrices and thereby generating a reference null distribution that controls for the network structure. Four steps are taken in an MRQAP analysis: (1) A classic linear or logistic regression is run with the values of the network matrix as a dependent variable and predictors of interest (eg, level of homophily) as independent variables. The regression coefficients are stored because only the estimated standard errors are affected by the violation of the independence assumption. (2) The network matrix that constitutes the dependent variable is permuted (in a simulation study, it was reported that permuting the dependent variable is the most conservative method to obtain statistical inference—other methods permute, for example, the independent variables or the residuals.15), which means that row and column orders are shuffled simultaneously. This way, a large set of networks (in our case, 2000) is created, where (on average) there should be no association between the dependent network and independent variables. (3) On each of these permuted dependent networks, a regression model is run with specifications identical to those in step 1. (4) In step 4, regression coefficients obtained in the analyses performed in step 3 (ie, the null distribution) are compared with the observed regression coefficients obtained in step 1. We can calculate, for example, the 97.5%/2.5% percentiles of the null distribution to obtain a measure similar to the 95% confidence interval (CI). Furthermore, we can compute a P value by assessing how many coefficients from the null distribution are larger or equally large than the observed value. For instance, a P value of 0.99 indicates that 99% of the coefficients based on permuted networks are smaller or equal to the observed estimates. The probability of observing a larger estimate under the null hypothesis is thus P = 0.01 (for further details, see Ref. 38). It is important to note that the level of analysis is the dyadic relationship and not (as usual) the individual. In that sense, we aimed at predicting the presence of a friendship nomination (RQ1 and RQ2) and the level of perceived positive friendship quality and negative friendship quality (RQ3) between 2 adolescents.

To take the multilevel/multigroup data structure into account, we applied a multigroup extension of the MRQAP framework. Similar to a multilevel analysis, all levels (ie, schools) were analyzed simultaneously.52 The only difference to the general MRQAP framework is that the permutations (step 3 mentioned earlier) were performed within each group (ie, within each school).10,19 This way, the baseline characteristics within each school (eg, the distribution of adolescents with pain) were preserved, allowing observations to be compared with alternative scenarios (ie, null distribution) in which ties would be randomly distributed within each school (and not among all schools). Because of this, the skewed distribution of the pain grades was also considered in the analytic procedure.

To examine RQs 1 and 2, a MRQAP model was specified with the presence of a tie (ie, friendship nomination) between 2 adolescents as the dependent variable and independent variables in the form of matrices representing predictors related to the RQs and control variables (school class and grade). These key predictors were pain sender, pain receiver, and pain similarity (Table 1). The term sender refers to the adolescent sending a tie, in other words, an adolescent who nominated peers (for an illustration, see Fig. 1). The pain sender effect thus captures the association between an adolescent's pain grade (see lightning flash on Fig. 1) and the tendency to nominate other peers (irrespective of their pain grade). Analogously, receiver refers to the adolescent receiving a tie, so an adolescent who was nominated as a friend by another adolescent (Fig. 1). In that sense, the pain of receiver effect captures the association between an adolescent's pain level (see lightning flash on Fig. 1) and the tendency to be nominated by others (irrespective of the adolescent's own pain level). The pain of receiver effect thus captures how pain is associated with being popular (RQ1). For the model testing RQ2, we included a variable indicating the level of pain similarity between adolescents (ie, the pain similarity effect). Pain similarity is defined as Psim= −1×|Psender−Preceiver|, where Psender and Preceiver represent the pain values of every dyad in the data set. Higher values represent a higher level of pain similarity. Furthermore, this model also included an interaction effect between pain of the sender and pain of the receiver, which assessed whether the effect of pain similarity was differently expressed at the lower or higher end of the pain scale (Table 1). Simply put, this interaction effect assessed whether the effect of pain similarity was different for adolescents with a lower pain grade vs adolescents with a higher pain grade. Higher values for this effect correspond with homophily at the higher end of the pain scale. In this statistical interaction, dyads in which both individuals experience a high pain grade (eg, 4) have a high value (ie, 4 × 4 = 16). Dyads with low pain grades (eg, 0 and 1), on the contrary, have a low value (0 × 1 = 0). In this way, we can also assess in a multivariate model if the level of homophily is related to the dyad's level of pain. A positive estimate of the Psender×Preceiver interaction would indicate that dyads with high pain levels are more likely to be friends than those with low pain levels (ie, that the homophily effect is stronger at the higher end of the scale than at the lower end of the scale).

Table 1 - Definition of network variables. Variables Definition Pain sender Pain of an adolescent who named peers Pain receiver Pain of an adolescent who is named by other adolescents Pain similarity Level of pain similarity between adolescents of a dyad Pain sender × pain receiver Indicates whether the effect of pain similarity was different for adolescents without pain and adolescents with a higher pain grade Sex sender Sex of an adolescent who named peers Sex receiver Sex of an adolescent who is named by other adolescents Sex same Adolescents of a dyad of the same sex
Figure 1.:

Illustration of sender and receiver effect. Peer 1 sends a tie to peer 2 (peer 1 nominates peer 2 as a friend). The (pain) sender effect refers to the association between (pain) characteristics of the sender of the tie (peer 1) with the appearance of a tie. The (pain) receiver effect refers to the association between (pain) characteristics of the receiver of a tie (peer 2) with the appearance of a tie. Lightning signs represent high levels of the peer's pain.

Multiple Regression Quadratic Assignment Procedure also allows us to analyze weighted networks, which means that the degree of positive and negative friendship qualities with peers can be included as continuous dependent measures in the analyses. To examine RQ3, a MRQAP model was fitted with the degree of positive friendship quality or negative friendship quality as the dependent network. In this model, the pain sender effect reflects the association between the adolescents' own level of pain and their self-perception of the dyadic relationship with their peers. The pain receiver effect reflects the association between the receiver's level of pain and the (sending) adolescents' self-perception of the dyadic relationship.

In the models testing RQs 1, 2, and 3, we further controlled for effects of sex (sex sender, sex receiver, same sex; Table 1), being in the same school grade (grade homophily), and being in the same class (class homophily). We further controlled for differences in baseline levels of nominations (model for RQ1 and RQ2) and baseline levels of positive/negative friendship quality (model for RQ3) between the 19 schools. For RQ1-3, estimates cannot be standardized because the standardization requires a standard error of the estimate. In the case of social networks, this standard error is biased because of the violation of the independence assumption, which states that observations are independent of each other. For reasons of consistency, we also reported unstandardized estimates for RQ4.

Finally, to examine RQ4, 2 standard multilevel models with observations of adolescents nested within schools were estimated.52 Two moderation models were tested, 1 with stress as a dependent variable and the other with depressive symptoms as a dependent variable. The main effects of pain grade, positive and negative friendship qualities, and sex were included in the model. The main interests in this analysis are the interaction effects between pain grades, on the one hand, and positive or negative friendship quality, on the other hand, with stress and depressive symptoms as dependent variables, testing RQ4. In this model, we also included 2-way interaction effects between sex and positive or negative friendship quality, and 3-way interactions between sex, pain grade, and positive or negative friendship quality. The analyses were performed in the statistical programming language R57 using the netglm18 and the lme46 package.

3. Results 3.1. Sociodemographic and pain characteristics

Table 2 summarizes the sociodemographic information and the distribution of pain grades in the final sample.

Table 2 - Descriptive statistics of the sample. Variables Percentage (N) Sex, girls 47.6 (1316) Age, 13 or 14 y 91.0 (2518) Immigrant background 24.0 (664) Low socioeconomic status 18.3 (506) Divorced parents 32.9 (909) Pain grade  0 48.2 (1335)  1 30.6 (848)  2 8.8 (243)  3 6.4 (178)  4 2.0 (54)
3.2. Social network characteristics

In total, 6489 social network nominations were recorded. Adolescents nominated a mean number of 2.34 (SD = 0.89) others. As an illustration, Figure 2 shows a network plot of 1 randomly selected school. The figure shows that few adolescents reported pain grade 3 or 4. Some were reciprocally connected to each other, meaning that they nominated each other as being part of the same peer group. In addition, Figure 2 shows isolated adolescents, who were not nominated by others as being part of a peer group and did not nominate others themselves. Network figures of all other schools can be found in the Supplementary Materials (available at https://links.lww.com/PAIN/B594).

Figure 2.:

Plot of network in school 8. Circles = female adolescents, rectangles = male adolescents. Color and size of nodes represent pain grade, and darker and bigger nodes represent higher pain grades (dark red = pain grade 4; yellow = pain grade 0). Color and width of ties represent higher positive (green) or higher negative (orange) friendship quality.

3.3. RQ1: are adolescents with higher pain grades less often nominated as being part of a peer group?

Table 3 summarizes the results of a multigroup MRQAP model with the presence of a friendship nomination as the dependent variable. The results showed that the effect of pain grade on nominations from peers (ie, pain receiver effect) was not significant (B = −0.003, P = 0.470), indicating that the number of nominations was not lower for adolescents with higher pain grades.

Table 3 - Results of Multiple Regression Quadratic Assignment Procedure model for research questions 1 and 2. B P E(Est.) Percentiles 2.5th 97.5th Intercept −11.658*** <0.001 −4.420 −4.553 −4.296 Pain sender 0.017 0.306 0.001 −0.059 0.063 Pain receiver −0.003 0.470 −0.001 −0.066 0.061 Pain similarity 0.075** 0.004 0.000 −0.056 0.057 Pain sender × pain receiver −0.002 0.472 0.000 −0.042 0.044 Sex sender (ref. female) 0.063* 0.011 −0.002 −0.052 0.051 Sex receiver −0.216*** <0.001 0.001 −0.051 0.054 Sex same 3.058*** <0.001 0.000 −0.051 0.052 Class same 3.049*** <0.001 −0.002 −0.065 0.063 Grade same 4.320*** <0.001 0.000 −0.049 0.052

*P < 0.05; **P ≤ 0.01; ***P ≤ 0.001. The reference category is female adolescents. E(Est) = Expected value of the estimate under the null distribution (ie, the mean value of regression coefficients in the permuted scenarios).

3.4. RQ2: do adolescents with similar pain grades name each other more often as being part of the same peer group?

Table 3 also summarizes the results regarding RQ2. These results indicated a significant effect of pain similarity on the presence of a friendship tie between adolescents (B = 0.075, P = 0.004), which means that adolescents with a similar pain grade named each other more often as being part of the same peer group. The interaction between pain sender and pain receiver was not significant, indicating that there was no difference in the effect of pain similarity across pain grades (B = −0.002, P = 0.472). In other words, adolescents with similar pain grades named each other more often regardless of their specific pain grade. So the homophily effect was equally driven by dyads of low and high pain grades. Figures 3 and 4 show these results visually. Figure 3 shows the predicted probabilities of friendship ties (nominations) with varying levels of pain grades of senders and receivers; in other words, it shows the chance of a friendship nomination considering both adolescents their pain grade. Figure 3 suggests that the probability of a friendship tie between adolescents was highest (dark red) when they reported similar pain grades, regardless of the level of their similar pain grade. Figure 4 shows the number of observed ties or friendship nominations (red dots) in comparison with a reference null distribution (white violin distribution) by pain similarity and adolescent's pain grade. The null distribution represents the number of nominations in a network in which ties would be randomly distributed. Figure 4 suggests that higher difference scores in pain grades were related to lower likelihood to nominate each other, so it was less likely that adolescents with different pain grades nominated each other (regardless of the specific pain grade).

Figure 3.: Predicted probabilities of friendship ties between adolescents with different pain grades. Darker red squares represent higher probabilities than darker blue squares. Predicted probabilities are computed based on the model estimates listed in Table 2.Figure 4.:

Number of ties by pain similarity and pain grade (<3 lower than pain grade 3; ≥3 equal to or higher than pain grade 3). Difference score 0 represents similar pain grades.

Furthermore, the MRQAP results in Table 3 indicated a significant effect of sex similarity, which means that same-sex adolescents named each other more often (B = 3.058, P < 0.001). Being in the same class (B = 3.049, P < 0.001) or school grade (B = 4.320, P < 0.001) was also highly significant, indicating that adolescents in the same class or school grade named each other more often as being part of the same peer group. Effect sizes of the latter 3 effects were very high, for instance, the likelihood of a same sex friendship was 21 times (= Exp (3.058)) larger than a non–same-sex friendship.

3.5. RQ3: do dyads with an adolescent with a higher pain grade report less positive friendship quality and more negative friendship quality? 3.5.1. Positive friendship quality

Table 4 summarizes the results of a multigroup MRQAP model with perceived positive friendship quality as the dependent variable. We focused on the direct effects of friendship quality because of a weak correlation (r = −0.15) between positive and negative friendship qualities. The effect of pain sender on positive friendship quality was significant (B = −0.059, P = 0.023), indicating that adolescents with a higher pain grade perceived the relationship with their friends as less positive regarding friendship quality than adolescents with lower pain grades. The effect of pain receiver was not significant (B = −0.035, P = 0.126), indicating that adolescents (who may or may not have pain) perceived relationships with friends with higher pain grades as equally positive as they perceived relationships with friends with lower pain grades.

Table 4 - Results of Multiple Regression Quadratic Assignment Procedure model for research question 3 with positive friendship quality as a dependent variable. B P E(Est.) Percentiles 2.5th 97.5th Intercept 5.686*** <0.001 4.013 3.887 4.144 Pain sender −0.059* 0.023 0.000 −0.058 0.058 Pain receiver −0.035 0.126 0.000 −0.059 0.058 Pain similarity −0.018 0.250 0.000 −0.052 0.052 Pain sender × pain receiver 0.025 0.121 0.000 −0.040 0.040 Sex sender (ref. female) −0.362*** <0.001 0.000 −0.053 0.053 Sex receiver −0.400*** <0.001 0.000 −0.050 0.048 Sex same −0.257*** <0.001 0.000 −0.048 0.051 Class same −0.171*** <0.001 0.001 −0.061 0.060 Grade same −0.089*** <0.001 0.000 −0.048 0.049

*P < 0.05; **P ≤ 0.01; ***P ≤ 0.001. The reference category is female adolescents. E(Est) = Expected value of the estimate under the null distribution (ie, the mean value of regression coefficients in the permuted scenarios).

Furthermore, the effect of pain similarity within dyads was not significant in this model (B = −0.018, P = 0.250), indicating that adolescents with similar pain grades did not report higher or lower levels of positive friendship quality. Finally, results also showed that male adolescents perceived the relationship as less positive than female adolescents (B = −0.362, P < 0.001) and that adolescents (irrespective of their own sex) perceived relationships with male adolescents as less positive than relationships with female adolescents (B = −0.400, P < 0.001). Peer relationships between adolescents of the same sex were perceived as less positive regarding friendship quality (B = −0.257, P < 0.001). Similarly, adolescents of the same class (B = −0.171, P < 0.001) or school grade (B = −0.089, P < 0.001) perceived the dyadic relationship as less positive. These tendencies may result from rare relationships that cut across the boundaries of sex, class, and grade but being perceived as highly positive by adolescents.

3.5.2. Negative friendship quality

Table 5 summarizes the results of a multigroup MRQAP model with perceived negative friendship quality as the dependent variable. The effect of pain sender on negative friendship quality was significant, indicating that adolescents with higher pain grades perceive the relationship with their friends as more negative (B = 0.134, P < 0.001). The effect of pain receiver was also significant (B = 0.064, P = 0.002), indicating that adolescents perceived relationships with peers with higher pain grades as more negative when compared with relationships with peers with lower pain grades. Of interest, the effect of pain sender is twice as large as the effect of pain receiver on negative friendship quality, indicating that higher pain grades have a larger effect on one's own perception of relationships than the perception from others.

Table 5 - Results of Multiple Regression Quadratic Assignment Procedure model for research question 3 with negative friendship quality as a dependent variable. B P E(Est.) Percentiles 2.5th 97.5th Intercept 1.360*** <0.001 1.591 1.500 1.680 Pain sender 0.134*** <0.001 0.000 −0.044 0.045