Adolescent Popularity as a Nonlinear,Two-Dimensional Construct: Convergent Results From Three Large Samples

Abstract

The goal of this study was to advance the conceptualization and measurement of adolescent popularity by exploring the commonly used composite score (popularity minus unpopularity). We used standardized peer nominations from 4,414 early adolescents (ages ≈ 12-14 years) from three samples collected in two countries. Popularity and unpopularity were strongly related, but not linearly; scatterplots of the two variables resembled an L-shaped right angle. Subsequent analyses indicated that either including popularity as a curvilinear term or including both popularity and unpopularity as separate terms explained significantly more variance in social and behavioral correlates than linear, bivariate analyses using popularity, unpopularity, or composite popularity. These results suggest that researchers studying adolescent popularity should either separate popularity and unpopularity or treat composite popularity as curvilinear.

Keywords

In the more than 20 years since Parkhurst and Hopmeyer (1998) presented empirical research distinguishing social preference from popularity, researchers have studied popularity extensively among youth (van den Berg, Lansu, & Cillessen, 2020). Unlike social preference, which indicates peers’ affective response to an adolescent, popularity represents adolescents’ position in the social hierarchy based primarily on social power, influence, and dominance (Marks, Cillessen, & Crick, 2012).

Previous research predominantly focused on high levels of popularity, either based on unipolar scores (i.e., “who is most popular”) or composite scores (i.e., “who is most popular” minus “who is least popular”; van den Berg et al., 2020). Unpopularity, in contrast, has received much less study. Qualitative investigations have long indicated that unpopular adolescents are low in social power and dominance, and are often victimized or marginalized by their high-status peers (e.g., Canaan, 1990; Eder, 1995), which is similar to the way that recent quantitative researchers have characterized the variable (e.g., Malamut, Mali, Schwartz, Hopmeyer, & Luo, 2017; Schwartz, Kelly, & Duong, 2013). Although low in power, unpopular adolescents may be second only to the popular adolescents in terms of social visibility, in part because popular adolescents may purposefully call negative attention to their unpopular peers as a way of maintaining their own position in the status hierarchy (Canaan, 1990; Eder, 1995).

Beyond the fact that unpopularity is a worthwhile facet of peer relations and status to investigate, the lack of empirical attention to unpopularity in the quantitative literature is problematic because it is often measured as part of assessing popularity. Some peer nomination research on popularity has used a single popularity item (most popular), but much of the literature has used a composite score, combining two items of popularity and unpopularity into a difference score (most popular minus least popular; van den Berg et al., 2020). No empirical study has yet examined whether popularity and unpopularity can be seen as opposite ends of a single continuum, whether the composite score is optimal from a psychometric and predictive perspective, or whether assessing popularity and unpopularity separately provides relevant information about adolescent status. The goal of this study was to address these issues.

Popularity, Unpopularity, and Composite Popularity

It is easy to understand why popularity and unpopularity might be combined into a single variable. Researchers in all areas of the social sciences have long viewed status as a single rank ordering of individuals from top to bottom (e.g., Benoit-Smullyan, 1944), and it seems intuitive that measuring adolescent popularity would require assessment of both ends of the status spectrum (i.e., both the top and the bottom of the hierarchy; Marks et al., 2012).

Beyond face validity, there are aspects of measurement that support a composite popularity score. Individual peer nomination items are often positively skewed and zero-inflated (see Gorman, Schwartz, Nakamoto, & Mayeux, 2011); a difference score between two skewed variables may have a more normal distribution. In addition, a composite variable based on multiple items may yield a more internally reliable measure than popularity or unpopularity separately (Babcock, Marks, Crick, & Cillessen, 2014). These are reasons to combine popularity and unpopularity into one score.

However, there are also reasons why popularity and unpopularity may not be polar opposites on one single spectrum. First, the two studies that have coded adolescent responses to open-ended questions have shown that popularity and unpopularity do not consistently relate to relational and behavioral variables in opposing directions. LaFontana and Cillessen (2002) concluded that teens seemed to view attractiveness and social connectedness as differentiators of popularity versus unpopularity, but that popular and unpopular youth were viewed similarly in terms of aggression, bullying, and prosocial behavior. Xie and colleagues (2006) found several differences between reports of popular versus unpopular seventh graders (e.g., prosocial behavior, social connection, and appearance) but no significant differences in reports of overt aggression, relational aggression, deviant behavior, or dominance.

Second, the few peer nomination studies that have analyzed popularity and unpopularity separately have consistently found negative correlations between them, which ranged from nonsignificant (Košir & Pečjak, 2005) to low (Gorman et al., 2011) to only moderate (Lease, Kennedy, & Axelrod, 2002; van den Berg & Cillessen, 2013). In addition, the social, relational, and behavioral correlates of popularity did not consistently mirror the correlates of unpopularity. For instance, some studies found that overt and relational aggression were positively related to both popularity and unpopularity (Gorman et al., 2011; Lease, Kennedy, & Axelrod, 2002). In contrast, van den Berg and Cillessen (2013) found that physical aggression was positively related to popularity, but either low-negative or nonsignificantly related to unpopularity, depending on the method of data collection (computerized vs. pencil-and-paper). Associations with rejection were inconsistent across these studies. Unpopularity and rejection were positively correlated, but the relation between popularity and rejection was negative and small (r = –.14; Lease, Kennedy, & Axelrod, 2002), nonsignificant (van den Berg & Cillessen, 2013), or even positive and small (Gorman et al., 2011). In looking at peer victimization, Gorman et al. found moderate (positive) associations of unpopularity with overt and relational victimization, but nonsignificant associations of popularity with both types of victimization. Van den Berg and Cillessen found positive associations of unpopularity with victimization and negative associations of popularity with victimization, but these associations were much stronger for unpopularity. Together, these results suggest that popularity and unpopularity are not simply opposites and have distinct social, relational, and behavioral correlates.

Based on their findings, Gorman and colleagues (2011) concluded “that unpopularity is distinct from low levels of popularity” (p. 215). Beyond treating popularity and unpopularity as continuous variables, this conclusion also has implications for categorizing adolescents into popularity categories. Some researchers have used composite popularity to identify subgroups of youth (Berger & Dijkstra, 2013; Closson, 2009; see also van den Berg, Burk, & Cillessen, 2015). Others have distinguished subgroups of popular and unpopular youth based on popularity nominations alone (Lease, Musgrove, & Axelrod, 2002; Rodkin & Ahn, 2009). If low popularity is not the same as high unpopularity, and vice versa, either of these categorizations may not be ideal. Therefore, an additional goal of this study was to determine whether single scores of popularity or unpopularity and composite popularity converge in the identification of popular and unpopular groups. In particular, it is worthwhile to know whether high-status and low-status youth, as identified by scores on popularity, correspond to high-status and low-status individuals, as identified by unpopularity, and whether either of these correspond to high-status and low-status individuals, as identified by the composite score of popularity.

The Current Study

The overall purpose of this study was to contribute to the further conceptualization and measurement of adolescent popularity by comparing the psychometric, statistical, and predictive properties of composite popularity with popularity and unpopularity treated as separate constructs. Using peer nomination data from three large samples, with more than 4,400 middle school participants, we addressed five research questions.

First, we examined the bivariate association between popularity and unpopularity. What is their overlap? Are there participants who are either high or low in both scores? To our knowledge, no previous study has looked beyond correlations to examine the association between separate scores for adolescent popularity and unpopularity in more detail.

Second, we investigated whether composite popularity had better psychometric properties (i.e., distribution and reliability) than popularity or unpopularity separately. Because the composite score combines two skewed variables, we expected its distribution to be more normal than popularity or unpopularity separately. Given that popularity separately already has high 1-year stability (Marks et al., 2012) and internal reliability (Marks, Babcock, Cillessen, & Crick, 2013), we did not expect composite popularity to improve upon it. It was unclear whether unpopularity would be as reliable as either popularity or composite popularity.

Third, we assessed whether categorizations based on popularity, unpopularity, and composite popularity converged when identifying high- and low-status groups.

Fourth, we determined whether popularity and unpopularity had divergent associations with other social and behavioral variables. We expected that correlations of popularity and unpopularity with other social or behavioral variables would not directly mirror each other. For example, in earlier research, aggression was more strongly related to popularity than unpopularity and victimization more strongly to unpopularity than popularity (Gorman et al., 2011; Lease, Kennedy, & Axelrod, 2002; van den Berg & Cillessen, 2013).

Fifth, we used multiple regression analyses to determine whether separating popularity and unpopularity nominations yielded greater predictive utility than composite popularity.

Method

Choices of Samples and Variables

To avoid the possibility that any observed similarities or differences between popularity and unpopularity would be sample-specific, we conducted analyses across three separate samples. We began with data from three large longitudinal studies. We chose to focus on the seventh-grade waves because, in all three studies, this age group was available and participation rates were high.¹ We chose to compare popularity and unpopularity with the specific social and behavioral variables used in this study for two reasons: (a) all of the social and behavioral variables were assessed in all three studies (although with varying numbers and wording of items), and (b) these variables are likely the most common peer nomination variables considered in the literature on adolescent popularity (see Cillessen & Marks, 2017). The longitudinal studies to draw from, the use of the seventh-grade waves from these studies, and the selection of specific social and behavioral variables to analyze were all determined a priori.

Sample 1

Sample 1 consisted of 598 adolescents representing all seventh-grade students from two middle schools (Ns = 282 and 316) in a midsized town in the northeastern region of the United States who were part of the Manchester Youth Study (see Marks et al., 2012) during the 1998-1999 wave. The sample was 48.8% female and included Caucasian (66.9%), Black/African American (19.4%), and Hispanic/Latino/Latina (11.9%) students. Of the full sample of 598 adolescents, 465 (77.8%) completed peer nominations and 541 (90.5%) were again assessed in eighth grade using the same measures.

In both grades, nominators were given a roster of all grade-mates for each item and were asked to circle the names of an unlimited number of peers who fit each criterion. The constructs used in the current analyses were popularity (participants asked to name “the people in your grade who are the most popular”), unpopularity (“the people in your grade who are the least popular”), friendship (“your best friends”), acceptance (“you like the most”), rejection (“you like the least”), overt aggression (“start fights, say mean things, and/or tease others”), relational aggression (two items; “ignore others or spread rumors about them when they are mad at them” and “try to keep others who they don’t like from being in their group”), prosocial behavior (two items; “cooperate, share, and help others” and “are leaders and good to have in charge”), overt victimization (two items; “get pushed and kicked by others” and “get picked on and teased”), and relational victimization (two items; “have lies, rumors, and mean things said about them” and “get left out of group activities when others are mad at them”).

Sample 2

Sample 2 consisted of 1,928 Dutch secondary school participants (971 seventh graders and 930 eighth graders; 49.9% female) who were part of the seventh wave of data collection (Spring/Summer 2011) of the Nijmegen Longitudinal Study (van den Berg et al., 2015). Participants were in 74 classrooms ranging from 15 to 31 students ( $\bar{X}$ = 26; SD = 3.70). Of the full sample, 1,812 students (94.0%) completed peer nomination measures on netbook computers.

Items were displayed on the top of the screen, followed by a list of all classmates; lists were randomized for each participant. Nominators could select an unlimited number of peers from the list. The constructs relevant to the current analysis were popularity, unpopularity, friendship, acceptance, rejection, overt aggression (three items), relational aggression (two items), prosocial behavior (three items), overt victimization (three items), and relational victimization (two items).

Sample 3

Sample 3 was drawn from the Kandinsky Longitudinal Study (see van den Berg, Burk, & Cillessen, 2019), a cohort sequential study that includes longitudinal data on annual cohorts of Dutch seventh graders at two schools. The current analyses included 1,888 seventh graders (50.6% female) from the six waves assessed between 2011 and 2016 (between 259 and 286 participants per cohort). Participants were in 69 classrooms, ranging in size from 16 to 32 students ( $\bar{X}$ = 27, SD = 2.68).

Of the full sample, 1,820 students (96.4%) completed peer nomination measures on netbook computers using the same procedure as in Sample 2. Constructs were the same as in Sample 2, with the exception that prosocial behavior was measured using a single item.

Furthermore, 1,331 participants (70.5%) of the seventh graders assessed in the 2011 through 2015 data collection waves were also nominees 1 year later in eighth grade using the same measures.

Data Preparation

For all three samples, scores for individual items were standardized within the reference groups (grades for Sample 1 and classrooms for Samples 2 and 3), and then summed for composite behavioral variables. Composite popularity was calculated based on the difference between standardized popularity and standardized unpopularity.

Results

Overview of Planned Analyses

We first conducted basic descriptive analyses focusing on popularity, unpopularity, and composite popularity, including the bivariate association between popularity and unpopularity, using a linear Pearson correlation, scatterplots between the two variables, skew, and kurtosis. We also calculated Cronbach’s alpha (within classrooms; Babcock et al., 2014) and 1-year stability correlations to see whether there were large statistical differences in the quality of data between composite popularity and the individual popular and unpopular variables. The 1-year stability of popularity, unpopularity, and composite popularity was calculated for Samples 1 and 3 using Pearson correlations.

In order to determine whether categorization of participants into popularity-based groups would differ depending on which variables were used (e.g., popularity, unpopularity, or composite), we conducted categorical analyses of youth high and low in popularity and unpopularity to see how well discrete levels of composite popularity correspond to the levels of most popular and least popular. We did so by splitting the samples into quartiles to ensure that identical numbers of participants were in each group for all three variables. Other criteria for splitting the groups, such as a fixed z score, could have led to differing interpretations across the three data sets, given differing shapes of distributions. In addition to examining absolute agreement, we calculated agreement using kappa to correct for chance-level agreement. We considered a kappa value of .25 (agreement that is 25% beyond what is possible by statistical chance) to be moderate agreement and a kappa value of .50 to be a high level of agreement. We divided each of the 4 × 4 quartile contingency tables into two 2 × 4 contingency tables (one table for the lower quantiles and one for the higher quantiles) in order to compare chance-corrected classification agreement in the first and second quartiles with the third and fourth quartiles (Howell, 2007).

Next, we correlated popularity and unpopularity with the social and behavioral variables. Both directions and magnitude were important. For example, if variable x correlates .20 with popularity, but –.80 with unpopularity, this suggests that popularity and unpopularity are not demonstrating opposite associations with x, but rather that they are distinct constructs. We used effect sizes (r²) to determine differences in the magnitude of the associations of popularity and unpopularity with the other variables. We determined a priori to highlight effect size differences that were .10 and .30 in magnitude.

Finally, we conducted a series of multiple regressions in which we predicted each criterion variable from composite popularity in Step 1, popularity and unpopularity in Step 2, and the interaction between popularity and unpopularity in Step 3.² The ΔR² values in Steps 2 and 3 were examined to determine whether social and behavioral variables were better predicted by separating popularity and unpopularity than by combining them into a composite variable.

Following these planned analyses, we also conducted two sets of post hoc analyses. First, we made additional scatterplots from external samples to replicate our findings from our three primary samples. Second, we examined curvilinear trends in regression analyses.

Descriptive Analyses

Bivariate correlations between popularity and unpopularity were r = –.20, –.43, and –.42 for Samples 1 to 3, respectively. The scatterplots (top row of Figure 1) showed that the associations between popularity and unpopularity were nonlinear in all three samples. Specifically, the associations between popularity and unpopularity resembled a right angle along (and just below) the x- and y-axes. This indicated that participants with z scores above 0 for popularity were unlikely to have z scores above 0 for unpopularity and vice versa. In fact, out of the 4,414 participants, only 17 (0.4%) had z scores above 0 for both popularity and unpopularity.

Figure 1.

Scatterplots of least popular z score on most popular z score for the three primary samples and eight replication samples.

Figure 1 also indicates that a low popularity score was not the same as a high unpopularity score. The negative linear correlations between the two masked their bivariate distribution, which was substantially heteroscedastic. The variation in unpopularity was much higher at low levels of popularity than at high levels of popularity, and the variation in popularity was much higher at low levels of unpopularity than at high levels of unpopularity.

Post Hoc Replication of the Association Between Popularity and Unpopularity

We explored whether the L-shaped association between popularity and unpopularity was also found in other samples. These data came from other grades assessed in the three studies and from an unconnected and new study (Mayeux, 2014). We examined scatterplots for the eighth- through 12th-grade data for Sample 1 (Ns ranged from 481 to 663), the ninth- through 10th-grade data for Sample 2 (N = 1,801), the eighth-grade data for Sample 3 (N = 1,368), and data from an unconnected data set of 530 ninth graders collected by Lara Mayeux in the Midwestern region of the United States (Mayeux, 2014). The scatterplots all demonstrated the same L-shaped associations (second and third row of Figure 1).

Psychometric Properties

As shown in Table 1, popularity and unpopularity were positively skewed and leptokurtic. In Samples 2 and 3, composite popularity was relatively normally distributed (skew and kurtosis close to 0). The highest levels of skewness and kurtosis for popularity and unpopularity were found in Sample 1, likely because the larger reference group in this sample (entire grade instead of classroom) allowed for a handful of extreme outliers for both popularity and unpopularity (see first scatterplot of Figure 1).³ Composite popularity was negatively skewed and leptokurtic; however, similar to Samples 2 and 3, the distribution of composite popularity was less skewed than for popularity or unpopularity.

Table 1.

Distribution and Reliability Metrics for Popularity, Unpopularity, and Composite Popularity Across Three Samples.

	Sample 1 (Manchester)			Sample 2 (Nijmegen)			Sample 3 (Kandinsky)
	Pop	Unpop	Comp pop	Pop	Unpop	Comp pop	Pop	Unpop	Comp pop
Skew	3.65	7.79	−1.27	1.51	1.83	−0.11	1.53	2.03	−0.17
Kurtosis	17.35	82.42	20.39	1.45	2.82	0.16	1.57	3.95	0.35
Cronbach’s α^a	.97	.91	.95	.94	.91	.93	.91	.93	.94
Stability (r)	.93	.89	.92	—^b	—	—	.75	.78	.81

Note. Pop = popularity; Unpop = unpopularity; Comp pop = composite popularity (i.e., Popularity – Unpopularity).

Cronbach’s alpha values represent the mean values across the two schools (Sample 1) or across all classrooms (Samples 2 and 3). Alphas for composite popularity were calculated using the “pasting method” described by Babcock et al. (2014).

1-year stability coefficients are not available from Sample 2 because the Nijmegen Longitudinal Study did not collect data in consecutive years.

Next, we computed the mean alphas across the reference groups in each sample (Table 1). Internal reliability was high (αs >.90) for both popularity and unpopularity across all three samples. The composite popularity variables did not yield higher values.⁴ In Sample 3, composite popularity was a bit more stable than popularity or unpopularity alone, but the differences were small (.06 and .03, respectively). In Sample 1, the stability of composite popularity was similar to that of popularity and unpopularity.

The overall pattern of results in Table 1 indicates that composite popularity was more normally distributed than popularity and unpopularity individually, but that composite popularity showed no major advantage in terms of internal reliability or stability.

Categorization of Highly Popular or Unpopular Youth

Table 2 contains statistics concerning absolute agreement between the z-score quartiles for popularity and unpopularity, and the quartiles for composite popularity. For high status, agreement represents youth who were in the same quartile for both popularity and composite popularity. For low status, agreement represents youth who were in opposite quartiles for least popular and composite popularity. The difference in direction is due to high nomination counts having opposite interpretations for the composite variable (e.g., high unpopularity nomination counts yield lower composite popularity). As seen in Table 2, there was less agreement in the bottom two quartiles than in the top two quartiles. The highest agreement in the bottom two quartiles was about the same as the lowest agreement in the top two quartiles, with near perfect agreement in the fourth quartiles. When investigating agreement using kappa, kappa ranged from .00 to .16 for the first two quartiles (0%-16% better classification agreement than chance), but from .59 to .85 for the upper two quartiles (59%-85% better than chance). Thus, the percent absolute agreement and kappa analyses both indicated that categorizations based on composite popularity agreed with popularity and unpopularity at the high end of their scores, but not at the low end.

Table 2.

Quartile Categorization Agreement Statistics Comparing Popularity and Unpopularity With Composite Popularity.

Quartile, popularity	Percentage in the same composite popularity quartile
Quartile, popularity	Sample 1	Sample 2	Sample 3
First	52%	60%	60%
Second	30%	47%	46%
Third	65%	88%	84%
Fourth	89%	98%	96%
Kappa for Quartiles 1 and 2	.01	.15	.14
Kappa for Quartiles 3 and 4	.61	.85	.81
Quartile, unpopularity	Percentage in the opposite composite popularity quartile
Quartile, unpopularity	Sample 1	Sample 2	Sample 3
First	48%	59%	57%
Second	31%	48%	45%
Third	60%	75%	75%
Fourth	95%	96%	97%
Kappa for Quartiles 1 and 2	.00	.16	.11
Kappa for Quartiles 3 and 4	.59	.73	.75

Intercorrelations

As shown in Table 3, correlations for popularity and unpopularity with the other social and behavioral variables were almost always in opposite directions, but the magnitude of effects indicated that popularity and unpopularity were not direct opposites. In all three samples, most correlations for popularity and unpopularity differed by an effect size of at least .10. These differences in magnitude were consistent across samples for several of the social and behavioral variables. Aggression tended to be more strongly related to popularity than unpopularity. Rejection and victimization were substantially more strongly correlated with unpopularity than popularity. Overall, the patterns of correlations for popularity and unpopularity were distinct.

Table 3.

Correlations Between Popularity Variables (Popularity, Unpopularity, and Composite Popularity) and Other Peer Nomination Variables.

	Sample 1			Sample 2			Sample 3
	Pop	Unpop	Comp pop	Pop	Unpop	Comp pop	Pop	Unpop	Comp pop
Most popular	—	−.20	.77	—	−.43	.85	—	−.42	.84
Least popular	−.20	—	−.77	−.43	—	−.85	−.42	—	−.84
Friendship	.65	−.29	.61	.32	−.63	.56	.31	−.50	.48
Acceptance	.75	−.31	.69	.26	−.48	.44	.30	−.47	.46
Rejection	.23	.44	−.14	−.02	.50	−.31	.03	.38	−.21
OA	.22	−.03	.16	.42	−.13	.32	.47	−.17	.38
RA	.65	−.09	.48	.59	−.28	.52	.58	−.27	.50
Prosoc	.62	−.17	.51	.34	−.40	.44	.01	−.22	.14
OV	−.13	.87	−.65	−.21	.65	−.51	−.09	.59	−.41
RV	−.02	.85	−.56	−.32	.86	−.70	−.27	.83	−.65

Note. Bold correlation coefficients indicate that, for the given sample, the effect size of the correlation between most popular and the target variable differs by at least 10% from the effect size of the correlation between least popular and the target variable (|r²_mpop − r²_lpop| > .10). Bold and underlined correlations differ by an effect size of at least 30% (|r²_mpop − r²_lpop| > .30). Pop = popularity; Unpop = unpopularity; Comp pop = composite popularity (i.e., Popularity – Unpopularity); OA = overt aggression; RA = relational aggression; Prosoc = prosocial behavior; OV = overt victimization; RV = relational victimization.

Predictive Utility of Combining versus Separating Popularity and Unpopularity

Finally, a series of multiple regression analyses were conducted to predict each criterion variable from composite popularity (Step 1), popularity and unpopularity (Step 2), and the Popularity × Unpopularity interaction (Step 3). Results for Steps 1 and 2 are displayed in Table 4 (note that Step 2 of this set of analyses is labeled as “Step 2a” in the Table). The R²s and the ΔR²s were statistically significant (p < .05) in all regressions. Given the large sample sizes in this study, these significant values were not consistently meaningful. For instance, in Sample 2, the ΔR² for Step 2 predicting prosocial behavior was only .003, with significance of p = .047. As such, we were most interested in the magnitude rather than the significance of ΔR²s in Step 2.

Table 4.

Regressions Predicting Social and Behavioral Variables From Composite Popularity (Step 1) and Either Popularity and Unpopularity (Step 2a) or Squared Composite Popularity (Step 2b).

	Sample 1			Sample 2			Sample 3
	Step 1: Comp popR²	Step 2a: Pop and unpopΔR²	Step 2b: Comp pop²ΔR²	Step 1: Comp popR²	Step 2a: Pop and unpopΔR²	Step 2b:Comp pop²ΔR²	Step 1:Comp popR²	Step 2a: Pop and unpopΔR²	Step 2b:Comp pop²ΔR²
Friendship	.37*	.08*	.06*	.31*	.08*	.07*	.23*	.03*	.02*
Acceptance	.47*	.12*	.08*	.19*	.04*	.04*	.21*	.02*	.02*
Rejection	.02*	.28*	.20*	.10*	.20*	.20*	.04*	.15*	.12*
OA	.02*	.02*	.01	.10*	.08*	.07*	.14*	.08*	.06*
RA	.23*	.20*	.13*	.27*	.09*	.07*	.25*	.08*	.07*
Prosoc	.26*	.12*	.13*	.19*	.00*	.00	.02*	.04*	.03*
OV	.42*	.34*	.33*	.26*	.17*	.18*	.16*	.21*	.21*
RV	.31*	.43*	.45*	.49*	.26*	.26*	.42*	.27*	.28*

Note. Two sets of multiple regression analyses are presented. Both sets of regressions use composite popularity in Step 1 of the analysis. The first set of regressions added popularity and unpopularity as separate predictors (Step 2a). The second set of regressions added a squared composite popularity term (Step 2b). Comp pop = composite popularity (i.e., Popularity – Unpopularity); Pop = popularity; Unpop = unpopularity; OA = overt aggression; RA = relational aggression; prosoc = prosocial behavior; OV = overt victimization; RV = relational victimization.

p < .05.

Adding popularity and unpopularity separately improved model fit to a nontrivial extent in most analyses, explaining at least 10% more variance than composite popularity for six criterion variables in Sample 1 and for three in Samples 2 and 3. Across samples, the prediction of rejection and both types of victimization benefited most from adding popularity and unpopularity separately, with predictions of aggression also benefiting.

Results of Step 3 showed that including the interaction between popularity and unpopularity did not improve the prediction of any social and relational variables. In five of the 24 cases, the interaction predicted only 1% to 2% additional variance over Step 2. In the remaining cases, the additional variance explained was less than 0.5%.

Post Hoc Analyses of Curvilinear Trends

Following the planned regression analyses, we visually analyzed the scatterplots of composite popularity with each of the social and behavioral variables (see Figure 2, for examples), and noted two things. First, the association between composite popularity and most of the criterion variables appeared to be curvilinear. Second, the inflection point of these curvilinear analyses was around a z score of 0 for composite popularity. We therefore conducted an additional series of multiple regression analyses to test whether the social and behavioral variables would be better predicted by treating composite popularity curvilinearly, rather than linearly. Composite popularity was the sole predictor in Step 1, and squared composite popularity was added in Step 2.

Figure 2.

Scatterplots and trends of three variables on most popular, least popular, and composite popularity z scores using Sample 2 data.

The results are displayed in Table 4, with the new results of Step 2 labeled “Step 2b.” Step 1 results were identical to those provided in the first series of regressions. The ΔR²s for Step 2 in these analyses were similar to the ΔR²s for Step 2 in the previous set of regression analyses. With a few exceptions in Sample 1, composite popularity as a curvilinear predictor had about the same predictive utility as popularity and unpopularity entered as separate variables. On average, adding popularity and unpopularity explained 2.5% more variance in the criterion variables in Sample 1 and less than 1% in Samples 2 and 3, compared with using a curvilinear composite popularity variable.

The fact that the inflection point of the curvilinear trends for composite popularity appeared to be at z scores of 0 suggested that the association between composite popularity and the social or behavioral variables might not truly be curvilinear. Rather, these trends could have been the result of popularity and unpopularity having different linear associations with the criterion variables. The shape of the associations between popularity and unpopularity (Figure 1) ensured that positive values of composite popularity were almost entirely the result of popularity scores and that negative values of composite popularity were almost entirely the result of unpopularity scores. Consequently, the curvilinear composite popularity trends might actually be the result of “pasting” two linear trends together.

Figure 2 shows how this “pasting” could occur.⁵ The three scatterplots in the left column show clear curvilinear trends for composite popularity. The two columns on the right show the linear trends for unpopularity and popularity. The displayed trends are only based on data greater than or equal to a z score of 0 for popularity and unpopularity, and the unpopularity graphs have been reversed to further improve the visual comparison. It is easy to see that the curvilinear associations for composite popularity may be the result of “pasting” the linear associations for popularity and unpopularity together.

The data below a z score of 0 for the scatterplots on the right half of Figure 2 are a concern. There was much noise at the low ends of some associations for both popularity and unpopularity, implying a floor effect. The possibility that popularity and unpopularity are opposite sides of the same construct would explain these floor effects—note that the floor effect in the association between popularity and relational victimization looks like the result of horizontally truncating the scatterplot of unpopularity and relational victimization. Similarly, the floor effect in the association between unpopularity and relational aggression looks like the result of horizontally truncating the scatterplot of popularity and relational aggression.

Summary of Results From Regressions and Scatterplots

These results do not provide evidence for the superiority of either separating popularity and unpopularity or using a composite popularity variable. However, the results of both sets of regression analyses in Table 4 (as illustrated by the scatterplots in Figure 2) do indicate that bivariate linear trends involving these status variables are not optimal for understanding how adolescent status relates to other variables. Composite popularity demonstrated curvilinear associations with most of the social and behavioral variables. As Figure 2 shows, however, linear associations between social and behavioral variables and popularity or unpopularity can be misleading, with unexplained variance in the low ends (i.e., the floor effect) potentially resulting in illusory correlations. These floor effects may be statistically controlled by including popularity and unpopularity as simultaneous predictors in analyses (as evidenced by the fact that Step 2a predicted about the same variance as Step 2b in the regressions in Table 4).

Discussion

The goal of this study was to examine the characteristics of a composite popularity variable and examine how it behaves in analyses compared with separate scores for popularity and unpopularity. Our findings were notably similar across three large samples from two countries, and the unexpected relation between popularity and unpopularity was replicated in eight additional samples. The results indicated that linear, bivariate analyses of popularity, unpopularity, or composite popularity may not optimally capture the popularity construct. These findings reinforce current thinking about adolescent popularity, but suggest ways that popularity can be better modeled in future research.

The Shape of Popularity

Popularity and unpopularity were not related in a linear fashion, but as an L-shaped right angle. Participants who received an above-average number of nominations for popularity tended to receive few or no nominations for unpopularity and vice versa. Fewer than 0.4% of participants were above the mean for both popularity and unpopularity, and there were zero bivariate outliers who were high on both variables. The consistency of this finding across the samples was remarkable.

The shape of the association between popularity and unpopularity is atypical in the statistical literature. Popularity and unpopularity cannot be said to be negatively correlated or related in any other linear sense—the low-to-moderate correlation coefficients previously observed between popularity and unpopularity may have been the result of the L-shaped distribution. At the same time, it is not justified to say that popularity and unpopularity are unrelated; on the contrary, they are highly related, but not in a sense that we would normally see for two-dimensional constructs.

Implications for Conceptualizing Popularity

These findings have implications for popularity research. On the one hand, previous research using a popularity item without an unpopularity item may have yielded illusory findings. On the other hand, research using composite popularity may have interpreted curvilinear relations from a linear perspective. These findings invite us to reexamine previous work.

Theoretically, understanding the L-shaped association between popularity and unpopularity adds to our understanding of adolescent popularity and provides a path for future research. Two key properties of adolescent popularity are that (a) it is based on a hierarchical continuum, and (b) it is entirely based on the social consensus of the peer group (Marks et al., 2012). The current findings strengthen these assertions. Although the association between popularity and unpopularity may be L-shaped, the current findings make it clear that each adolescent is at a distinct position on the popularity spectrum between extreme popularity and extreme unpopularity. Similarly, the fact that no participants scored high in both popularity and unpopularity supports the notion that popularity is based on a broad consensus of the peer group. Nominators overwhelmingly agreed upon which peers were popular and which were unpopular.⁶

This study also highlights the need of research on unpopularity as a facet of adolescent status. The ways that adolescents think about popularity has been a topic of considerable research (e.g., Adler & Adler, 1998; Cillessen & Marks, 2011; Milner, 2004; Parkhurst & Hopmeyer, 1998). Yet unpopularity has never been fully defined or explored, despite being commonly measured. Thus, we have an operational definition of the construct, but we lack a conceptual definition, a theoretical framework, or a robust quantitative and qualitative literature that allows for interpretation and prediction.

Conclusions Regarding the Composite Popularity Variable

An implicit assumption of the difference score for composite popularity is that the underlying construct is a linear continuum with popularity at one end and unpopularity at the other. The current findings support the notion that the popularity construct is a spectrum ranging from popularity to unpopularity, but contradict the assumption that the spectrum is linear. The difference score for composite popularity can be seen as essentially flattening the two-dimensional associations by pivoting one axis along the fulcrum at the right angle itself until the distribution is one-dimensional.

The nonlinear association between popularity and unpopularity is different from the linear association between acceptance and rejection. Notably, there was no group of participants in this study who would be “controversial” using popularity and unpopularity (i.e., high in both constructs; Coie, Dodge, & Coppotelli, 1982). This may explain why even studies that have considered popularity in relation to the five preference-based groups identified by Coie and colleagues (1982) have split adolescents into three popularity-based groups (popular, average, or unpopular; LaFontana & Cillessen, 1999; Parkhurst & Hopmeyer, 1998).

In addition, whereas acceptance and rejection overlap somewhat within the social preference score (e.g., an “average” individual may have any number of combinations of acceptance and rejection nominations), composite popularity seems to more clearly separate its component scores. Scores below 0 on composite popularity were almost entirely based on unpopularity, and scores above 0 were almost entirely based on popularity. This provides a robust explanation for the curvilinear associations found in this study between composite popularity and the social and behavioral variables. In many cases across the three samples, the association between composite popularity and the criterion variables was qualitatively different above and below a composite popularity score of 0.

This study is not the first to find curvilinear associations between composite popularity and social and behavioral variables—they have been reported by Prinstein and Cillessen (2003) and subsequent studies (e.g., Prinstein, Choukas-Bradley, Helms, Brechwald, & Rancourt, 2011; Stoltz, Cillessen, van den Berg, & Gommans, 2016; Walcott, Upton, Bolen, & Brown, 2008). However, this study is the first to suggest that curvilinear associations between composite popularity and other variables may result from a nonlinear association between popularity and unpopularity. Our regression results indicated that curvilinear associations were the rule, not the exception, when using composite popularity as a predictor.

Separating Popularity and Unpopularity

Our results provided some evidence that popularity and unpopularity are distinct constructs. Bivariate correlations showed that popularity and unpopularity were related to different variables. For instance, popularity was strongly associated with aggression, whereas unpopularity was associated with victimization. In addition, the first set of regression analyses showed that separating popularity and unpopularity explained more variance in behavioral criteria than a linear composite popularity variable.

Our results also indicate that popularity and unpopularity are not separate or separable constructs. Although the associations between composite popularity and criterion variables differed above and below the 0 point for composite popularity (i.e., they were nonlinear), they were consistently continuous. That is, the y-intercepts were the same for both popularity and unpopularity. This is reflected in the fact that composite popularity as a curvilinear variable accounted for about the same variance as separating popularity and unpopularity.

Additional evidence for not separating popularity and unpopularity is in the scatterplots in Figure 2. In the scatterplots using unpopularity and popularity as predictors, we included linear regression lines only for scores above z = 0. This is because, in some cases, there was large error variance in the negative range—examples are most clearly visible in the scatterplot of relational victimization by popularity and of relational aggression by unpopularity. This error variance resembles a floor effect for the predictor, and the results suggest that the error variance associated with the floor effect for unpopularity is the effect of popularity and vice versa. The floor effects explain why bivariate correlations often seem to be in opposite directions for popularity and unpopularity. For example, although the association between popularity and relational victimization was mostly flat in Sample 2, the bivariate correlation between popularity and relational victimization was significant and negative due to the error variance introduced by the strong positive association between unpopularity and relational victimization (see Figure 2). These floor effects for popularity and unpopularity were eliminated when they were combined into one score or when both were entered simultaneously in a regression (i.e., the variance associated with floor effects was accounted for when both variables were controlled).

As part of our evaluation of popularity, unpopularity, and composite popularity in this study, we conducted a series of bivariate correlations. As subsequent analyses progressed, however, it became clear that none of the correlations were accurate representations of the associations. The association between composite popularity and the social or behavioral variables was curvilinear in most cases, making linear conclusions inappropriate. In addition, it became clear that some of the significant correlations that included either popularity or unpopularity were illusory, based on the discussed link between the two constructs.

Conclusions for Popularity Research: Recommendations for Analysis and Terminology

The determinations that popularity and unpopularity are (a) inseparable, and (b) related in a nonlinear fashion lead to the conclusion that popularity research should avoid bivariate linear analyses using popularity, unpopularity, or composite popularity variables. The current findings suggest three analytical strategies. The first is to use composite popularity and treat it as a curvilinear variable. The second is to use popularity and unpopularity as simultaneous predictors of social and behavioral outcomes. This study provides insufficient evidence to support one of these strategies over the other. On the one hand, our regression analyses indicated that separating popularity and unpopularity explained slightly more variance in criterion variables than the curvilinear term for composite popularity. On the other hand, the differences were small, and it can be argued that using a single curvilinear term is more parsimonious than two terms that are strongly connected. In addition, the popularity and unpopularity variables violated assumptions of normality more than composite popularity.

The third strategy for future research is with regard to groups. The composite popularity variable was accurate in identifying the most popular and unpopular youth but inaccurate in identifying youth who were low in popularity or low in unpopularity. Therefore, researchers should use both popularity and unpopularity or composite popularity when creating subgroups of high- and low-status youth.

In terms of terminology, researchers should first and foremost be intentional, and as specific as possible, in writing about these variables. Although we recommend against analyzing popularity and unpopularity in isolation, when referring to results using only a “most popular” item, for example, researchers should refer to high or low popularity (or high/low in nominations of most popular, which is comparable to the way that many researchers refer to “liked most” and “liked least” nominations). Researchers should never refer to individuals low in popularity as “unpopular.” A variable based on subtracting “least popular” nominations from “most popular” nominations should be referred to as “composite popularity.”

Future Research

Much of our existing knowledge regarding adolescent popularity is based on linear, bivariate analyses. The current findings indicate that these associations may be more complex. First, although the current results were correlational, they suggest that the processes governing popularity and unpopularity are distinct. Youth may lose popularity without gaining unpopularity. By the same token, if the pivot point in the right angle between popularity and unpopularity indicates average levels of status, then variables associated with gaining or losing status may be different for youth who are more popular than average than for youth who are more unpopular than average (see van den Berg et al., 2019). Future research will need to explore this possibility.

Individual differences that might affect the processes of gaining popularity and unpopularity should also be considered. For example, the importance of gender and gender norms in gaining and losing status has been long established in both the qualitative (e.g., Adler & Adler, 1998) and quantitative (e.g., Mayeux & Kleiser, 2020) literatures. It is possible that gender, along with variables like race/ethnicity or SES, may further interact with behaviors in determining how changes in popularity correspond to changes in unpopularity, or lack thereof.

Contextual differences in processes of popularity and unpopularity also warrant investigation. Further cross-cultural comparisons regarding such processes are vital and particularly research in non-Western cultures. Classroom-level variables are similarly worth considering. Correlates of status have been found to differ depending on classroom norms (e.g., Boor-Klip et al., 2017), for example. Reference group size may also be important; although the relations between popularity and unpopularity as variables were similar in Sample 1’s grade-wide context versus the other two samples’ classroom-level reference groups, the processes guiding status may differ depending on the size of the reference group (e.g., being extremely popular or unpopular at the grade level may mean different things than at the classroom level).

More broadly, additional research may investigate particular constellations of traits that may be associated with popularity and/or unpopularity. For example, in the current analyses, aggression was positively associated with popularity but not unpopularity, whereas victimization was positively associated with unpopularity but not popularity. This leads to the question of where bully-victims, who are identified both by aggression and victimization, stand in the popularity hierarchy. In our data, no youth were high in both popularity and unpopularity. Perhaps aggression and victimization cancel each other out, leading bully-victims to be average in popularity? Or is the combination of aggression and victimization associated with either above- or below-average popularity?

As part of such future investigations of characteristics and behaviors associated with popularity and unpopularity, it is important that researchers consider correlates assessed by methods other than peer nominations. Although peer nominations are appropriate for collecting data on the affective reactions and interpersonal behaviors assessed in this study (Cillessen & Marks, 2017), many are best measured using other methods. Our understanding of popularity and unpopularity may benefit from minimizing shared methods variance. Furthermore, given that different reporters may provide different behavior and status-relevant information (e.g., as when peer nominations of victimization do not correspond to self-reports of victimization; Dawes, Chen, Farmer, & Hamm, 2017; Graham & Juvonen, 1998), it is vital that researchers differentiate relations between status and peer reputations of behavior, self-perceptions of behavior, and assessments of behavior from other parties (e.g., teachers, parents).

Limitations

One limitation of this study is that, although the large majority of our analyses were planned, some of our conclusions were based on subjective interpretations of the results, rather than on a priori criteria such as null hypothesis significance tests. Certainly, a visual analysis of scatterplots was the starting point for the post hoc analyses that contributed to our conclusions regarding popularity. Despite this, we believe that our findings generate a variety of new hypotheses to be tested in further research.

In addition, it is unclear whether the nonlinear association between popularity and unpopularity is specific to peer nominations. Although nominations are the most common method of assessing popularity and unpopularity, further research will need to determine whether curvilinear analyses are also found when data are collected using (for example) peer ratings.

Conclusion

It is possible that the observed association between popularity and unpopularity is one way the two are related, without being the only way they are ever related. Based on three data sets with over 4,400 participants (and replicated in additional samples), we believe that treating popularity as a curvilinear construct does provide an appropriate basis for future research and allows us to uncover more properties of the popularity construct itself. In addition, it is important for future researchers to avoid solely conducting bivariate, linear analyses involving popularity, regardless of the way it is quantified. If our framework is inaccurate or context-specific, researchers will find that linear analyses explain associations to composite popularity better than curvilinear analyses or that most popular and least popular scores can be separated without confounding each other. If our framework for popularity is accurate, however, researchers studying adolescent popularity will need to adjust their analyses to further develop our understanding of adolescent status, its concurrent correlates, and later outcomes.

Footnotes

Authors’ Note

The views and discussions presented in this research are not necessarily the official views of the Association for Materials Protection and Performance.

Acknowledgments

The authors would like to thank Dr. Matthew Findley, who initially suggested that any methodological investigation should begin with scatterplots, and Dr. Lara Mayeux, who permitted us to use her data for some of the post hoc analyses described in this article.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Data collection for Sample 1 was supported by a grant from the University of Connecticut Research Foundation to the first author. Data collection for Sample 2 was supported by Grant 9534900 from Stichting Kinderpostzegels Nederland to the third author. Data collection for Sample 3 was supported by a grant from the Behavioural Science Institute at Radboud University, Nijmegen to the third and last authors.

ORCID iDs

Peter E. L. Marks

Yvonne H. M. van den Berg

Notes

Author Biographies

Peter E. L. Marks is associate professor of developmental psychology at Austin College in Sherman, Texas, USA. His research focuses on adolescent popularity and peer nomination methodology.

Ben Babcock is the senior psychometrician and certification developer at The Association for Materials Protection and Performance. His research interests include psychometrics, measurement, and the statistical properties of peer nomination measures.

Yvonne H. M. van den Berg is assistant professor of developmental psychology in the Behavioural Science Institute (BSI) at Radboud University, Nijmegen, The Netherlands. Her research interests include child and adolescents peer relationships and classroom social dynamics.

Rob Gommans is a PhD candidate in the Department of Interdisciplinary Social Science at Utrecht University and the research data officer at the Behavioural Science Institute at Radboud University. His research focuses on peer influence processes (e.g., conformity) among children and adolescents, and developmental methodology and statistical analyses, specifically concerning sociometric measurements, social network analyses, and group composition research.

Antonius H. N. Cillessen is professor and chair of developmental psychology and director of the Behavioural Science Institute at Radboud University in Nijmegen, The Netherlands. His research interests include peer relations in childhood and adolescence, social behavior, peer influence, and quantitative methods for developmental research (sociometry, social networks, and longitudinal data analysis). He is the coeditor of Popularity in the Peer System (2011, Guilford Press).

References

Adler

P. A.

Adler

(1998). Peer power: Preadolescent culture and identity. New Brunswick, NJ: Rutgers University Press.

Babcock

Marks

P. E. L.

Crick

N. R.

Cillessen

A. H. N.

(2014). Limited nomination reliability using single- and multiple-item measures. Social Development, 23, 518-536. https://doi.org/10.1111/sode.12056

Benoit-Smullyan

(1944). Status, status types, and status interrelations. American Sociological Review, 9, 151-161. https://doi.org/10.2307/2086307

Berger

Dijkstra

J. K.

(2013). Competition, envy, or snobbism? How popularity and friendships shape antipathy networks of adolescents. Journal of Research on Adolescence, 23, 586-595. https://doi.org/10.1111/jora.12048

Boor-Klip

H. J.

Segers

Hendrickx

M. M. H. G.

Cillessen

A. H. N.

(2017). The moderating role of classroom descriptive norms in the association of student behavior with social preference and popularity. Journal of Early Adolescence, 37, 387-413. https://doi.org/10.1177/0272431615609158

Canaan

(1990). Passing notes and telling jokes: Gendered strategies among American middle school teenagers. In Ginsburg

Tsing

(Eds.), Uncertain terms: Negotiating gender in American culture (pp. 215-231). Boston, MA: Beacon Press.

Cillessen

A. H. N.

Marks

P. E. L.

(2011). Conceptualizing and measuring popularity. In Cillessen

A. H. N.

Schwartz

Mayeux

(Eds.), Popularity in the peer system (pp. 25-56). New York, NY: Guildford Press.

Cillessen

A. H. N.

Marks

P. E. L.

(2017). Methodological choices in peer nomination research. In Marks

P. E. L.

Cillessen

A. H. N.

(Eds.), New directions in peer nomination methodology. New directions for child and adolescent development (Vol. 157, pp. 21-44). San Francisco, CA: Jossey-Bass. https://doi.org/10.1002/cad.20206

Closson

L. M.

(2009). Status and gender differences in early adolescents’ descriptions of popularity. Social Development, 18(2), 412-426.

10.

Coie

J. D.

Dodge

K. A.

Coppotelli

(1982). Dimensions and types of status: A cross-age perspective. Developmental Psychology, 18, 557-570.

11.

Dawes

Chen

Farmer

Hamm

(2017). Self- and peer-identified victims in late childhood: Differences in perceptions of the school ecology. Journal of Youth and Adolescence, 46, 2273-2288. https://doi.org/10.1007/s10964-017-0688-2

12.

Eder

(1995). School talk: Gender and adolescent culture. New Brunswick, NJ: Rutgers University Press.

13.

Gorman

A. H.

Schwartz

Nakamoto

Mayeux

(2011). Unpopularity and disliking among peers: Partially distinct dimensions of adolescents’ social experiences. Journal of Applied Developmental Psychology, 32, 208-217.

14.

Graham

Juvonen

(1998). Self-blame and peer victimization in middle school: An attributional analysis. Developmental Psychology, 34, 587-599.

15.

Howell

D. C.

(2007). Statistical methods for psychology (6th ed.). Belmont, CA: Thomson Wadsworth.

16.

Košir

Pečjak

(2005). Sociometry as a method for investigating peer relationships: What does it actually measure? Educational Research, 47, 127-144.

17.

LaFontana

K. M.

Cillessen

A. H. N.

(1999). Children’s interpersonal perceptions as a function of sociometric and peer perceived popularity. The Journal of Genetic Psychology, 160, 225-241. https://doi.org/10.1080/00221329909595394

18.

LaFontana

K. M.

Cillessen

A. H. N.

(2002). Children’s perceptions of popular and unpopular peers: A multimethod assessment. Developmental Psychology, 38, 635-647. https://doi.org/10.1037/0012-1649.38.5.635

19.

Larkin

R. W.

(1979). Suburban youth in cultural crisis. New York, NY: Oxford University Press.

20.

Lease

A. M.

Kennedy

C. A.

Axelrod

J. L.

(2002). Children’s social constructions of popularity. Social Development, 11, 87-109. https://doi.org/10.1111/1467-9507.00188

21.

Lease

A. M.

Musgrove

K. T.

Axelrod

J. L.

(2002). Dimensions of social status in preadolescent peer groups: Likability, perceived popularity, and social dominance. Social Development, 11, 508-533. https://doi.org/10.1111/1467-9507.00213

22.

Malamut

S. T.

Mali

L. V.

Schwartz

Hopmeyer

Luo

(2017). Depressive symptoms as a predictor of social difficulties in a gang-impacted context. Journal of Applied Developmental Psychology, 51, 4-11.

23.

Marks

P. E. L.

Babcock

Cillessen

A. H. N.

Crick

N. R.

(2013). The effects of participation rate on the internal reliability of peer nomination measures. Social Development, 22, 609-622. https://doi.org/10.1111/j.1467-9507.2012.00661.x

24.

Marks

P. E. L.

Cillessen

A. H. N.

Crick

N. R.

(2012). Popularity contagion among adolescents. Social Development, 21, 501-521.

25.

Mayeux

(2014). Understanding popularity and relational aggression in adolescence: The role of social dominance orientation. Social Development, 23, 502-517.

26.

Mayeux

Kleiser

(2020). A gender prototypicality theory of adolescent peer popularity. Adolescnt Research Review, 5, 295-306. https://doi.org/10.1007/s40894-019-00123-z

27.

Milner

(2004). Freaks, geeks, and cool kids. New York, NY: Routledge.

28.

Parkhurst

J. T.

Hopmeyer

(1998). Sociometric popularity and peer-perceived popularity: Two distinct dimensions of peer status. Journal of Early Adolescence, 18, 135-144. https://doi.org/10.1177/0272431698018002001

29.

Prinstein

M. J.

Choukas-Bradley

S. C.

Helms

S. W.

Brechwald

W. A.

Rancourt

. (2011). High peer popularity longitudinally predicts adolescent health risk behavior, or does it? An examination of linear and quadratic associations. Journal of Pediatric Psychology, 36, 980-990. https://doi.org/10.1093/jpepsy/jsr053

30.

Prinstein

M. J.

Cillessen

A. H. N.

(2003). Forms and functions of adolescent peer aggression associated with high levels of peer status. Merrill-Palmer Quarterly, 49, 310-342. https://doi.org/10.1353/mpq.2003.0015

31.

Rodkin

P. C.

Ahn

(2009). Social networks derived from affiliations and friendships, multi-informant and self-reports: Stability, concordance, placement of aggressive and unpopular children, and centrality. Social Development, 18, 556-576. https://doi.org/10.1111/j.1467-9507.2008.00505.x

32.

Schwartz

Kelly

Duong

(2013). Do academically-engaged adolescents experience social sanctions from the peer group? Journal of Youth and Adolescence, 42, 1319-1330.

33.

Stoltz

Cillessen

A. H. N.

van den Berg

Y. H. M.

Gommans

(2016). Popularity differentially predicts reactive and proactive aggression in early adolescence. Aggressive Behavior, 42, 29-40. https://doi.org/10.1002/ab.21603

34.

van den Berg

Y. H. M.

Burk

W. J.

Cillessen

A. H. N

. (2015). Identifying subtypes of peer status by combining popularity and preference: A cohort-sequential approach. Journal of Early Adolescence, 35, 1108-1137. https://doi.org/10.1177/0272431614554704

35.

van den Berg

Y. H. M.

Burk

W. J.

Cillessen

A. H. N

. (2019). The functions of aggression in gaining, maintaining, and losing popularity during adolescence: A multiple-cohort design. Developmental Psychology, 55, 2159-2168.

36.

van den Berg

Y. H. M.

Cillessen

A. H. N

. (2013). Computerized sociometric and peer assessment: An empirical and practical evaluation. International Journal of Behavioral Development, 37, 68-76.

37.

van den Berg

Y. H. M.

Lansu

T. A. M.

Cillessen

A. H. N

. (2020). Preference and popularity as distinct forms of status: A meta-analytic review of 20 years of research. Journal of Adolescence, 84, 78-95.

38.

Walcott

C. M.

Upton

Bolen

L. M.

Brown

M. B.

(2008). Associations between peer-perceived status and aggression in young adolescents. Psychology in the Schools, 45, 550-561. https://doi.org/10.1002/pits.20323

39.

Xie

Boucher

S. M.

Hutchins

B. C.

Cairns

B. D.

(2006). What makes a girl (or a boy) popular (or unpopular)? African American children’s perceptions and developmental differences. Developmental Psychology, 42, 599-612. https://doi.org/10.1037/0012-1649.42.4.599