Strengthening Sociometric Prediction: Scientific Advances in the Assessment of Children's

Peer Relations
Author(s): Melissa E. DeRosier and James M. Thomas
Source: Child Development, Vol. 74, No. 5 (Sep. - Oct., 2003), pp. 1379-1392
Published by: Wiley on behalf of the Society for Research in Child Development
Stable URL: http://www.jstor.org/stable/3696184
Accessed: 11-07-2016 05:16 UTC

REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/3696184?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact [email protected].

Wiley, Society for Research in Child Development are collaborating with JSTOR to digitize,
preserve and extend access to Child Development

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
 Child Development, September/October 2003, Volume 75, Number 5, Pages 1379-1392

Strengthening Sociometric Prediction: Scientific Advances in the

Assessment of Children's Peer Relations

Melissa E. DeRosier and James M. Thomas

This study assessed the strength of sociometric classification in the prediction of concurrent sociobehavioral
adjustment. Differential adjustment for subgroups of unclassified children were also examined. Participants
were 881 fifth graders (ages 9 to 12). Classification strength (CS) and unclassified subgroups were determined
through newly developed algorithms. CS added significantly to the prediction of all areas of adjustment. For
example, highly rejected children were at extreme risk for victimization whereas highly controversial children
were most likely to be bullies and relationally aggressive. Unclassified subgroups were found to exhibit
adjustment problems mirroring those of their extreme status group counterparts. Findings support that
increasing the sensitivity of sociometric measurement results in both greater predictive strength and enhanced
understanding of underlying social processes.

The study of children's peer relations has a long and assigning children to social status groups based on
notable history dating back to Moreno's (1933, 1934) the pattern of positive (i.e., liked most) and negative
early work in the 1930s. Since that date, research has (i.e., liked least) nominations received from peers.
supported Moreno's contention that a key dimen- Using a standard score procedure, children receiving
sion of interpersonal relationships lies along the extreme scores for nominations are assigned to one
continuum of acceptance versus repulsion (i.e., of four extreme sociometric status groups: (a)
rejection). A primary tool for assessing this inter- popular (many positive, few negative nominations),
personal dimension within the peer group at school (b) controversial (many positive and negative nomi-
is sociometric methodology (see Cillessen & Bu- nations), (c) rejected (few positive, many negative
kowski, 2000, for review). Researchers administer a nominations), and (d) neglected (few positive or
peer nomination technique so that a set of children negative nominations). A fifth group-average-
(typically a class or grade at school) can nominate consists of children falling in the midrange for both
peers who match specific descriptions (e.g., children nominations. Any child falling between the average
you like the most). By comparing scores across the group and one of the five extreme groups is
nominating group, a child's social functioning in a considered unclassified.
particular area can be evaluated relative to that of Over the past several decades, abundant support
the broader peer group. has been provided for the predictive and concurrent
Coie, Dodge, and Coppotelli (1982) extended validity of Coie et al.'s (1982) sociometric classifica-
sociometric assessment by developing a system for tion groups as well as for the stability of status over
time and across settings (e.g., Cillessen, Bukowski, &
Haselager, 2000). The rejected status group has
Melissa E. DeRosier, 3-C Institute for Social Development, and proven a reliable and valid category of peer
Frank Porter Graham Child Development Center, University of problems (see Coie, 1990, for review). The insidious,
North Carolina at Chapel Hill; James M. Thomas, 3-C Institute for damaging influence of peer rejection (i.e., active
Social Development. dislike, avoidance, and exclusion by peers) on
The authors are indebted to the staff and students of the Wake
children's school-based functioning has been repeat-
County Public School System for their participation. Special
thanks to Ron Anderson for his support in the implementation edly demonstrated (see Parker, Rubin, Price, &
of this research. Robert Terry's consultation regarding traditional DeRosier, 1995, for review). Research has shown
sociometric categorization is much appreciated. The second that the predictive strength of rejection is not simply
author's efforts in the development of the statistical models and an artifact of other associated problems, such as
computer software (i.e., SCAN) used for this study are greatly
aggression or low socioeconomic status (SES) but
appreciated. SCAN software implementing the methodology
described in this article is the subject of a pending U.S. patent rather serves as an independent and unique con-
application. tributor to future adjustment problems (DeRosier,
Correspondence concerning this article should be addressed to
Melissa E. DeRosier, 3-C Institute for Social Development, 1903
Harrison Avenue, Suite 101, Cary, NC 27513. Electronic mail may ? 2003 by the Society for Research in Child Development, Inc.
be sent to [email protected]. All rights reserved. 0009-3920/2003/7405-0010

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
1380 DeRosier and Thomas

Kupersmidt, & Patterson, 1994; Kupersmidt & rejected status groups (i.e., in the rejected quadrant
DeRosier, in press). plotting positive and negative nominations), that
Although peer rejection has been found to be a child would have a rejected bias. It is highly likely
significant and independent predictor of a range of that unclassified children would differ on adjust-
negative outcomes, a large portion of unexplained ment depending on whether they are leaning toward
variance remains (McDougall, Hymel, Vaillancourt, popular versus rejected, for example. In this study, it
& Mercer, 2001). Reported effect sizes predicting was expected that significant differences in adjust-
adjustment problems 3 years later range from 5% ment would be found for the unclassified group
(internalizing) to 25% (externalizing and academic) according to subgroup biases. In particular, it was
explained variance (Coie et al., 1982; DeRosier et al., expected that unclassified children with a rejected
1994). Clearly, there may be methods that could bias would show the highest level of sociobehavioral
strengthen the prediction of maladjustment related adjustment problems.
to peer problems. The goal of the current study was Another central aspect of the current study was
to extend Coie et al.'s (1982) status group classifica- the examination of gender differences. Research with
tion system to gain greater sensitivity regarding a elementary students has shown that females tend to
child's place within a status group and to examine place greater emphasis on close, intimate friendships
whether greater sensitivity of measurement resulted whereas males tend to place greater emphasis on
in additional predictive strength for concurrent acceptance within larger social networks of peers
adjustment problems. (Buhrmester & Furman, 1987; Rutter, 1983). The
Two strategies were used to examine whether impact of rejection on sociobehavioral adjustment
measuring within-group differences resulted in may be tempered by the presence of at least one
enhanced prediction. First, for the four extreme positive friendship (Asher, Parker, & Walker, 1996).
status groups (i.e., popular, controversial, rejected, Given that sociometric status assesses acceptance
neglected), the strength of a child's classification was within the peer group, not individual friendships,
measured to capture the degree to which a child fell and that friendship has greater salience for females,
within their assigned status group. In this way, it was expected that the impact of classification
children who fell on the border of a status group strength (CS) on adjustment may be less significant
could be compared with those who fell at the for females than for males.
extreme or in the midrange. Given that children
who experience chronic rejection are at significantly
greater risk for maladjustment that those who Method

experience less frequent rejection (DeRosier et al.,

Participants
1994), it was expected that children who are more
extremely rejected would demonstrate poorer School selection. Ten public elementary schools
school-based adjustment than children who fall just from the Wake County Public School System
within the border of rejected status. It was also (WCPSS) in North Carolina participated in the study.
expected that incorporating the strength of a child's The WCPSS serves a large and diverse population of
sociometric classification into the analytic model more than 37,000 students in 74 elementary schools.
would significantly increase the prediction of con- The first author held an informational session with
current sociobehavioral adjustment. school personnel in the preceding spring to describe
The second strategy examined within-group the project and solicit participation. The 10 schools
variation for the unclassified group. Children who were accepted into the study (on a first-come, first-
fall within the unclassified group are typically served basis) to ensure an adequate sample size to
dropped from studies and receive no further con- examine subtypes of the extreme social status
sideration. This trend is unfortunate because it groups. Fifth grade was targeted for this study
eliminates the ability to examine children who may based on research suggesting the particular salience
be in transition from one status group to another and of peer relations at this developmental period (e.g.,
the processes that may promote such transitions. Furman & Buhrmester, 1992) and the reported
Children within the unclassified group fall some- escalation of peer problems, such as bullying, in late
where in between the average group and one of the elementary years (e.g., Ericson, 2001).
four extreme status groups. It is possible to assess an Parental consent. All fifth graders within the 10
unclassified child's bias toward an extreme status schools were eligible to participate. In September,
group. In other words, if an unclassified child's parent information letters describing the research
scores placed him or her between the average and project were mailed directly to the home of each fifth

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
 Strengthening Sociometric Prediction 1381

grade student within the 10 schools. Parents re- colleagues (Coie & Dodge, 1988; Coie et al., 1982).
turned the signed consent form via self-addressed, Liked most (LM) and liked least (LL) nominations
stamped envelope. Of the total pool of 962 students, were totaled and transformed to standard scores
parental consent was obtained for 881 students (M =0, SD = 1) by school (i.e., ZLM and ZLL,
(92%). respectively). Social preference (SP) was calculated
Sample characteristics. The sample was evenly by subtracting the ZLL score from the ZLM score.
distributed across genders (50.8% male, 49.2% Social impact (SI) was calculated by adding the ZLM
female) with a mean age of 10.6 years (range = 9.8- and ZLL scores. SP and SI were then restandardized
12.9). The approximate racial distribution was 73% within each school (i.e., ZSP and Zsi). Table 1 lists the
White, 20% African American, and 7% other racial criteria used to assign children to each social status
minorities. The SES of the sample ranged from group. The following percentage of the sample was
lower-class to upper-middle-class families. Across assigned to each group: popular = 14% (n = 120),
the 10 schools, the average number of classrooms controversial = 6% (n = 54), rejected = 14% (n = 123),
was four (range = 3-6) and the average class size neglected = 10% (n = 90), average = 14% (n = 119),
was 24 students (range = 18-26). and unclassified = 42% (n = 375). This distribution
closely matches that typically found using Coie's
Materials (1982) sociometric classification system.

For each item, children were presented with a CS

roster of all the children in their grade at their school
and were asked to nominate all the peers across the Extending beyond the traditional sociometric
grade who matched specific social acceptance and classification method (Coie et al., 1982), a new
social behavior descriptions (Coie et al., 1982; Crick algorithm was used to provide a richer under-
& Grotpeter, 1995). To assess social status, children standing of children's social position within their
responded to two standard positive and negative grade. As detailed in Table 1, to be classified within a
nominations: (a) children they like the most (liked particular social status group, a child must meet a
most), and (b) children they like the least (liked distinct set of conditions. For example, for a child to
least). To assess social behavior, children nominated be classified as popular, all three of the following
peers that fit each of the following five behavioral conditions must be met:
descriptions (Coie et al., 1982): (a) act like a bully a
1. standardized SP score greater than 1 (Zsp> 1),
lot (bully); (b) get picked on or called names a lot
2. standardized LM greater than 0 (ZLM > 0 ), and
(victim); (c) do mean things indirectly, such as 3. standardized LL less than 0 (ZLL <0 ).
spread rumors or organize exclusion (relational
aggression); (d) play or work alone a lot (with- If a child fails to meet just one of these conditions,
drawn), and (e) are often a good leader (leader). he or she will not be classified as popular. To
Gradewide nominations are the standard for late measure the strength of a child's classification (i.e.,
elementary school grades (Cillessen, Bukowski, & the degree to which a child falls within a status
Haselager, 2000). In this study, fifth graders had group), we must measure the extent to which each of
sufficient contact and familiarity with students in the conditions that determine that classification are
other classrooms, so that gradewide nominations met. For the popular classification, we must calcu-
across classrooms were used. Unlimited nomina- late the extent to which each child's (Zsp> 1), (ZLM
tions were also used in this study. Research indicates > 0 ), and (ZLL < 0 ). In fact, each condition that must
that unlimited nominations decrease error variance be met for all six of the social status groups is
and improve the stability and reliability in the expressed as a certain standardized score (e.g., Zsp,
measurement of children's peer relations (Terry, ZLL, ZLM) being greater than or less than a particular
2000). numeric cutoff (e.g., < 1, > 0, > .5). Each condition is
For each social behavior item, the number of expressed in terms of a cumulative density function
nominations a child received was summed and (CDF) that aggregates the area under the normal
standardized (M = 0, SD= 1) within the school. curve up to a particular value (an example is
Thus, children's social behavior was estimated available in Hart, 1968). The estimate of the extent
relative to the level of that behavior within their to which a given child's score (Zx) exceeds a given
peer group (Coie et al., 1982). cutoff (C) is given by CDF(Zx-C). Conversely, the
Children were assigned to sociometric status extent to which a given child's score falls short of a
groups using the algorithm developed by Coie and cutoff is given by 1-CDF(C-Zx).

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
1382 DeRosier and Thomas

0
The probability that the population mean is at least
Z,1 Criterion one below Child A's score is given by CDF(Zsp - 1)
Student A Cut-off = +1 0 or CDF(0), which equals 50%. Similarly, the prob-
ability that the population mean is at least one below
Child B's score is given by CDF(2-1) = CDF(1),
which equals 84%.
Using the CDF formulas, we can estimate the
nF o-50 1
likelihood that a given child's scores satisfy each of
the constituent rules used to classify children into
Student B
each sociometric status group. The resulting base set
Z . \2.0 of estimates is combined to calculate the degree to
which each child meets the set of criteria for each
2
social status classification. Specifically, as Table 1
1\I details, this conditional probability, P(Statusx), is
\; \\ F\\I -:84 ': derived as the product of the probabilities for each
status group's constituent rules (see the Appendix
Figure 1. Example calculations of probabilities for different social for additional information on the statistical proce-
preference scores based on the cumulative density function (CDF). dures).
Once computed for a given child, these six
conditional probabilities are normalized, or scaled,
An example of calculations of probabilities for relative to each other by dividing each by the sum of
students with different SP scores is displayed in all six (see Han & Kamber, 2000, p. 105, for more
Figure 1. One of the conditions that must be met for information regarding data transformation via scal-
popular status is that Zsp must be greater than or ing procedures). Scaling guarantees that the prob-
equal to 1. Another way of stating this condition is abilities are relative to each other and that they sum
that the population mean must be at least one less to 1, enabling characterizations of their relative
than the student's Zsp score. In Figure 1, Child A has influence on a given child's classification. These
a Zsp score of 1 (i.e., equal to the cutoff) and Child B scaled scores are referred to as the status group bias
has a Zsp score of 2 (i.e., 1 greater than the cutoff). (or SGB) scores. The highest of these six SGB scores,

Table 1

Algorithm for Determining Status Group Bias

Social status group Criteria Status group bias formula

Popular ZSP > 1 P(Popular) = CDF(Zsp - 1) x CDF(ZLM) x (1 - CDF(ZLL))

ZLM >0
ZLL < 0
Rejected Zsp < - 1 P(Rejected) = CDF( - 1 - Zsp) x (1 - CDF(ZLM)) X CDF(ZLL)
ZLM < 0
ZLL > 0
Neglected Zsi< - 1 P(Neglected) = CDF(- 1 - Zsi) x (1 - CDF(ZLM)) * (1 - CDF(ZLL))
ZLM <0
ZLL < 0

Controversial Zsi > 1 P(Controversial) = CDF(ZsI - 1) x CDF(ZLM) x CDF(ZLL)

ZLM > 0
ZLL > 0
Average - 0.5 < Zsp < 0.5 P(Average) = (1 - CDF(Zsp - .5)) x (CDF(.5+Zsp) x (1 - CDF(ZsI - .5) x CDF (.5+Zsi))
- 0.5 < Zs5 < 0.5
Unclassified - 1<Zsi < 1 P(Unclassified) = [(1 - CDF(Zsp - 1)) x (1 - CDF( - 1 - Zsp)) x (1 - CDF(ZsI - 1)) x
- 1 <Zsp < 1 (1 - CDF( - 1 - Zsi))] - P(Average)
Not average

Note. CDF = cumulative density function; P = conditional probability; Zsp = standardized social preference score; ZsI = standardized
status impact score; ZLM = standardized liked most score; ZLL = standardized liked least score. The formulas presented here are the core
estimates. However, the final calculation of status group bias (SGB) scores additionally used a multivariate approach to correct for
intercorrelated criteria. For complete details regarding the full SGB algorithms, please contact the authors.

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
 Strengthening Sociometric Prediction 1383

which corresponds to the child's designated status One staff member lead the group through the peer
group classification (e.g. SGB(Rejected) for a rejected nominations using a standardized classroom admin-
child) is referred to as the classification bias (CB) istration script. Two additional staff members
score. This is a continuous variable with a value supervised completion and answered children's
between 0 and 1, which represents the relative questions, as needed. To make a nomination,
strength of a particular child's bias toward their children were instructed to circle the number next
designated sociometric classification. To identify to the name of all the students on the roster who met
those children who fall just within the border versus that description. As recommended by researchers
in the midrange versus at the extreme of the popular, who study the administration of sociometric meth-
rejected, neglected, and controversial status groups, odology (e.g., Bell-Dolan & Wessler, 1994), several
the CB score is converted to a three-level categorical strategies were used to ensure privacy and quiet
variable CS. (It is not useful to compute CS scores for during administration as well as confidentiality
average and unclassified groups because there is not following administration. These strategies included
enough statistical width to these groups to achieve the use of folders to form a shield around ques-
differentiation.) Specifically, a CS score of 1 (i.e., low tionnaires for private responding, use of distractors
strength) is assigned to students whose CB score after administration (i.e., interesting prizes), discuss-
falls at the 25th percentile or below. Children with ing the importance of confidentiality with the class,
CB scores in the moderate range (i.e., between the and active monitoring during administration. Tea-
25th and 75th percentiles) are assigned a CS score of chers were not present in the classroom during
2 (i.e., moderate strength). Children are assigned a administration but rather supervised any nonparti-
CS score of 3 (i.e., high strength) if their CB score cipating students in the media center.
falls at the 75th percentile or above. Thus, a child
with a CB score of 1 falls just within the border of a Results
status group whereas a child with a CB score of 3
falls in the extremity of the status group.
Overview
Unclassified Bias
The results are divided into four sections. First,
To determine an unclassified child's bias toward the distribution of CS scores by status group and
one of the four extreme status groups, the ZLM and gender are explored. Second, children's sociobehav-
ZLL scores were plotted (with the intersection at a ioral adjustment is examined as a function of
mean of 0 for each score). A score of UP (unclassified sociometric status and CS, and the relative contribu-
popular), UR (unclassified rejected), UC (unclassi- tion of including CS in the model is examined. Third,
fied controversial), or UN (unclassified neglected) gender differences in the patterns of results are
was assigned to unclassified children to indicate in tested. Fourth, differences in adjustment as a
which quadrant of the plot they fell. UP falls in the function of unclassified bias and gender are exam-
upper left quadrant (i.e., ZLM in the positive range ined.
and ZLL in the negative range). UR falls in the lower
right quadrant (i.e., ZLM in the negative range and
CS by Status Group and Gender
ZLL in the positive range). UC falls in the upper right
quadrant (i.e., ZLM in the positive range and ZLL in Table 2 displays the distribution of CS scores
the positive range). UN falls in the lower left across status groups and genders. It shows that 50%
quadrant (i.e., ZLM in the negative range and ZLL of the sample fell within the moderate level of CS for
in the negative range). each status group and approximately 25% fell within
the low- and high-CS groups. Chi-square analyses
were performed to test for gender differences in the
Design and Procedure
distribution of males and females across the three CS
In November, pencil-and-paper questionnaires levels for the total sample and by status group. There
were group administered to children within class- were no significant differences in these distributions.
rooms by three trained staff members. Children were Adjustment as a Function of Sociometric Status and CS
presented with one roster for each sociometric
question (i.e., seven pages). For each school, all To examine children's sociobehavioral adjustment
children in the fifth grade at that school were as a function of sociometric status and CS, two
included on each roster (range = 70-160 students). multivariate analyses of variance (MANOVAs) were

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
1384 DeRosier and Thomas

Table 2

Classification Strength Distributions by Status Group and Gender

Classification strength score

Status group 1 (Low) 2 (Moderate) 3 (High)

Popular 30 62 28
Males 12 (40%) Females 18 (60%) Males 33 (53%) Females 29 (47%) Males 14 (50%) Females 14 (50%)
Rejected 28 63 32
Males 17 (61%) Females 11 (39%) Males 45 (70%) Females 18 (30%) Males 26 (81%) Females 6 (19%)
Controversial 13 29 12
Males 9 (69%) Females 4 (31%) Males 18 (62%) Females 11 (38%) Males 7 (58%) Females 5 (42%)
Neglected 22 45 23
Males 11 (50%) Females 11 (50%) Males 14 (31%) Females 31 (69%) Males 8 (35%) Females 15 (65%)
Total 93 (24%) 199 (51%) 95 (25%)
Males 49 (53%) Females 44 (47%) Males 110 (55%) Females 89 (45%) Males 55 (58%) Females 40 (42%)

conducted with sociobehavioral adjustment areas as rejected group was lowest on leadership and the
the dependent variables. First, a one-way MANOVA Neglected group was the next lowest.
with status as the independent variable was con- For the hierarchical model, both the main effect
ducted. Second, CS was added to the model both as for CS, F(10, 742) = 8.42, p< .0001, and the interaction
a main effect and as an interaction with status. By effect of CS x Status, F(30, 1486) = 4.96, p<.0001,
testing the hierarchical model, it was possible -to were significant at the multivariate level. (The main
determine whether CS added to the prediction effects for status were the same as those presented
beyond that of status alone and, if significant, to for the previous model and therefore will not be
establish the relative contribution of including CS in repeated here.) The univariate ANOVAs revealed
the model. that the CS main effect held for three areas of
In the MANOVA for the first model, status was adjustment: bully, F(2, 375) = 3.01, p<.05; victim,
significant at the multivariate level, F(15, F(2, 375) = 28.50, p <.0001; and withdrawn, F(2,
1047) = 45.48, p <.0001. The univariate analyses of 375) = 4.73, p <.01. Table 3 displays the means,
variance (ANOVAs) revealed that this effect held for standard deviations, and SNK post hoc mean
each area of adjustment: bully, F(3, 383) = 33.16; comparison results for each main effect. For all three
victim, F(3, 383) = 64.75; relational aggression, F(3, areas, the high CS level was associated with
383) = 49.88; withdrawn, F(3, 383) = 16.74; and lea- significantly more negative adjustment than the
der, F(3, 383) = 133.65; all ps < .0001. Table 3 displays low CS levels. For withdrawn and victim, the high
the means, standard deviations, and Student-New- CS level was also significantly higher than the
man-Keuls (SNK) post hoc mean comparison results moderate CS level.
for each area. Peers reported that the rejected and The status and CS main effects were qualified by a
controversial groups exhibited equally high levels of significant interaction effect for all five areas of
bullying behavior. However, only the rejected group adjustment: bully, F(6, 375) = 2.15, p <.05; victim,
was seen as experiencing a high level of victimiza- F(6, 375) = 14.83, p <.0001; relational aggression, F(6,
tion by peers. Peers reported that the controversial 375) = 2.94, p <.01; withdrawn, F(6, 375) = 4.91,
group exhibited the highest level of relational p <.0001; leader, F(6, 375) = 2.65, p <.05. Figures 2
aggression compared with all other groups. The through 6 display this interaction effect for each area
rejected group was seen as next highest on relational of adjustment. To determine the status group(s) for
aggression compared with the popular and ne- which the interaction held and the direction of effect,
glected groups. For withdrawn behavior, the rejected post hoc mean comparison tests were performed
and neglected groups were seen as equally high, comparing CS levels for each status group. The
whereas the popular and controversial groups were significant differences revealed through these
seen as equally low. For leadership, each group was post hoc tests are displayed in each figure
significantly different from all others. Peers reported (if no letters are displayed, there were no significant
that the popular group was the highest on leader- differences across the means for that status
ship followed by the controversial group. The group).

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
 Strengthening Sociometric Prediction 1385

Table 3

Means, Standard Deviations, and Explained Variance for Adjustment as a Function of Sociometric Status and Classification Strength

Sociometric status group CS level R2

Area Popular Rejected Controversial Neglected Low Moderate High Status Status + CS %Increase

Bullying - 0.36b 0.73a 0.79a - 0.40b - 0.13b 0.14ab 0.26a 20 23 15%

(0.45) (1.51) (1.71) (0.31) (0.88) (1.12) (1.51)
Victimization - 0.42b 1.36a - 0.08b - 0.30b - 0.04b 0.02b 0.89a 33 52 58%
(0.25) (1.87) (0.63) (0.43) (0.68) (0.94) (2.15)
Relational aggression - 0.43c 0.65b 1.04a - 0.53c - 0.02a 0.07a 0.25a 28 32 14%
(0.54) (1.36) (1.49) (0.40) (0.89) (1.11) (1.53)
Withdrawn - 0.25b 0.52a - 0.46b 0.30a - 0.13b 0.07b 0.36a 11 20 82%
(.60) (1.48) (0.37) (1.14) (0.83) (1.05) (1.44)
Leader 1.05a - 0.83d 0.66b - 0.55c - 0.03a 0.02a 0.11a 51 54 6%
(1.05) (0.32) (1.29) (0.49) (0.89) (1.18) (1.33)

Note. Different letters within a row signify significantly different means. CS = classification strength. The >Status+CS = R2 is the
combined contribution of the status main effect, the CS main effect, and the CS x Status interaction. The unique R2 contribution of the CS
main effect alone was 1%, 4%, 1%, 1%, and 1% across the five adjustment areas, respectively.

CS Level a 0 CS Level
I WLo: S1.5 OLow
0o OModerate
() 1 OModerate
I~ ab1
b
b ab o,

Ob

o a:
r b0
b -0

o..

Popua r Rejected Controversial Neglected Popular Rejected Controversial Neglected

Social Status Group Social Status Group

Figure 2. Bullying as a function of social status and classification Figure 3. Relational aggression as a function of social status and
strength (CS). Letters are displayed for status groups in which classification strength (CS). Letters are displayed for status groups
there were significant differences across CS categories. Subgroups in which there were significant differences across CS categories.
with different letters were significantly different from one another Subgroups with different letters were significantly different from
for that status group. one another for that status group.

For bullying, significant differences were found rejected children significantly more relationally
for the rejected and controversial groups. Low-level aggressive than low-level rejected children. Highly
rejected children were not seen as bullies by peers. controversial children showed the highest level of
However, moderately and highly rejected children relational aggression, being significantly higher than
showed significant elevations in peer-reported bul- low- or moderate-level controversial children as well
lying behavior. For controversial children, all three as significantly higher than all other CS levels across
levels were elevated for bullying. However, high- the other status groups.
level controversial children were significantly higher For victimization, significant differences were
than low-level controversial children, with the only found for the rejected group. Although both
moderate-level children falling between the other low- and moderate-level rejected children were seen
two levels. Relational aggression showed a similar as victims by peers, high-level rejected children were
pattern. Again, low-level rejected children were not significantly elevated for experiencing peer victimi-
seen as highly relationally aggressive by peers. zation. Across all status subgroups, highly rejected
Moderately and highly rejected children showed children were seen as the most victimized compared
elevations for relational aggression with highly with all other groups.

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
1386 DeRosier and Thomas

For withdrawn behavior, significant differences and relational aggression and less withdrawn
were found for the rejected and neglected groups. behavior than females. The multivariate interaction
Low- and moderate-level rejected children were not for Status x Gender was significant, F(15, 983) = 2.84,
seen as highly withdrawn by peers, whereas high- p<.001, and was present at the univariate level for
level rejected children were significantly elevated for bully, F(3, 360) = 4.33, p <.01, and victim, F(3,
withdrawn behavior. For the neglected group, low- 360) = 3.34, p < .05. SNK post hoc mean comparisons
level neglected children were not seen as withdrawn within status groups revealed that the gender
by peers and were significantly lower than the difference for bullying was only present for con-
highly neglected children. The moderately neglected troversial and rejected males who were found to
group fell in the middle but did not significantly bully significantly more than their female counter-
differ from the other two groups. parts. For victimization, rejected males were re-
For leadership, significant differences were found ported to experience greater victimization than
for the popular and controversial groups. Although rejected females. All other status groups showed
all three levels of each group were seen as leaders by no gender differences in victimization.
peers, highly popular and highly controversial Gender was not found to interact with CS either at
children were seen as significantly greater leaders the two-way (CS x Gender) or the three-way (CS x
compared with their respective low- and moderate- Status x Gender) interaction level. Therefore, the
level children (moderate-level popular children fell patterns of results for CS discussed in the previous
in between the other two levels). section were not found to vary by gender.

Relative Contribution of Including CS Unclassified Bias

To examine the relative contribution of including Children within the unclassified group were
information regarding CS, the percentage of ex- assigned to one of four bias subgroups according
plained variance (R2) for each area of adjustment to the procedure outlined earlier: popular bias
was compared across the two predictive models (n = 127), rejected bias (n = 91), controversial bias
outlined earlier. Table 2 displays the R2 for the model (n = 48), and neglected bias, (n = 109). Chi-square
including status as the only independent factor and analyses revealed that there were no gender differ-
for the model including CS in addition to status (as a ences for any bias subgroup. To test whether the bias
main effect and in interaction with status). The subgroups differed significantly in adjustment and
percentage increase in explained variance as a to examine differences in the pattern of results by
function of including CS is provided in Table 3. gender, a 4 (bias) x 2 (gender) MANOVA was
Across the five areas, CS information contributed performed predicting adjustment. All three effects
between 3% and 19% additional explained variance. were significant at the multivariate level: bias, F(15,
In effect, including CS in the model resulted in an 1003) = 16.25, p <.0001; gender, F(5, 363) = 10.47,
increase of between 6% (leader) and 82% (with- p<.0001; and Bias x Gender interaction, F(15,
drawn) in R2 compared with including status 1003) = 2.50, p<.01. The univariate main effects for
information alone. gender are discussed here because they directly
mirror those reported in the previous section.
The main effect for bias was present at the
Gender Differences in the Pattern of Results
univariate level for all five areas of adjustment:
To test whether the pattern of results for CS and bully, F(3, 367) = 27.00, p <.0001; victim, F(3,
the CS x Status interaction varied by gender, a 4 367) = 15.30, p<.0001; relational aggression, F(3,
(status) x 3 (CS)x 2 (gender) MANOVA was con- 367) = 32.12, p<.0001; withdrawn, F(3, 367) = 3.01,
ducted predicting adjustment. All main and inter- p <.05; and leader, F(3, 367) = 42.51, p <.0001. Table 4
action effects were included. However, to avoid displays the means, standard deviations, and SNK
redundancy, only effects involving gender are dis- post hoc mean comparison results for each area.
cussed here. The multivariate main effect for gender Peers reported that the UR and UC groups exhibited
was significant, F(5, 356) = 9.37, p<.0001, and was equally high levels of bullying behavior. However,
present at the univariate level for bully, F(1, only the UR group was seen as experiencing a high
360) = 12.08, p<.001; relational aggression, F(1, level of victimization by peers. Peers reported that
360) = 5.05, p <.05; and withdrawn, F(1, the UC and UR groups exhibited the highest level of
360) = 14.46, p<.001. SNK post hoc mean compar- relational aggression and the UP and UN groups
isons revealed that males exhibited more bullying were equally low. For withdrawn behavior, the UR

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
 Strengthening Sociometric Prediction 1387

Table 4

Means and Standard Deviations for Adjustment as a Function of Unclassified Bias and Gender

Unclassified bias group Average group

Area UP (n = 127) UR (n = 91) UC (n = 48) UN (n = 109) (n = 119)
Bullying - 0.36b 0.33a 0.29a - 0.34b - 0.05
(0.40) (1.08) (1.03) (0.35) (0.85)
Males - 0.27bl 0.57al 0.56al - 0.34bl 0.06
(0.43) (1.22) (1.15) (0.33) (0.89)
Females - 0.44al - 0.07a2 - 0.08a2 - 0.33al - 0.15
(0.34) (0.64) (0.70) (0.37) (0.81)
Victimization - 0.38c 0.15a - 0.21bc - 0.17b -0.11
(0.31) (0.87) (0.40) (0.56) (0.47)
Relational aggression - 0.41b 0.31a 0.48a - 0.35b 0.03
(0.48) (1.02) (1.02) (0.46) (0.89)
Withdrawn - 0.08ab 0.05a - 0.35b 0.11a - 0.09
(0.93) (0.92) (0.36) (1.14) (0.83)

Leader 0.53a - 0.53d 0.16b - 0.28c - 0.10

(0.96) (0.48) (0.86) (0.58) (0.73)

Note. Different letters within a row signify significantly different means. Different numbers within a column signify significantly different
means across genders. Standard deviations appear in parentheses. UP = unclassified with popular bias; UR = unclassified with rejected
bias; UC = unclassified with controversial bias; UN = unclassified with neglected bias.

and UN groups were seen as significantly higher 1988; Coie et al., 1982) have consistently provided
than the UC group; UP was not significantly valid and reliable measurement of children's social
different from any other group. For leadership, each status. However, the criterion values used for
group was significantly different from all others. inclusion into a sociometric status group are arbi-
Peers reported that the UP group was the highest, trary in nature (Cillessen & Bukowski, 2000). These
followed by the UC group, then the UN group, and values are not naturally occurring phenomenon but
finally the UR group. It is important to note that this rather are imposed by the researcher as reasonable
pattern of results directly mirrors the status group and logical statistical cutoffs. The cutoffs used for the
differences displayed in Table 2. The difference in current system of sociometric classification have
the patterns lies in the magnitude of many of the been so successful in their predictive value that, in
corresponding scores. effect, efforts to extend beyond them have been
The main effect for bully was qualified by a largely stymied. This study represents a significant
significant interaction effect with gender, F(3, statistical and conceptual step forward in our
367) = 4.89, p<.01. The within-gender means, stand- understanding of sociometric status and children's
ard deviations, and SNK post hoc mean comparison peer relationships.
results are included in Table 4. For females, there were Consistent with past research, findings from this
no subgroup differences. However, for males, the UR study support status group differences in patterns of
and UC groups were significantly higher than the UP social behavior (e.g., Coie et al., 1982). Clearly,
or UN groups. In addition, there were no gender sociometric status demarcates useful and informa-
differences for the UP and UN groups. However, for tive social differences. However, as expected, the
the UR and UC groups, males were significantly strength of a child's sociometric classification was
higher on bullying behavior than were females. found to add significantly to the prediction of
concurrent sociobehavioral adjustment. In other
Discussion words, capturing the degree to which a child fell
within a sociometric status group provided greater
For many years across countless studies, research understanding of his or her social behavior. These
has supported the value and predictive validity of significant within-group differences signify that all
using sociometric classification for children's peer children within a given status group should not be
relationships (Cillessen & Bukowski, 2000; Coie, considered equivalent and that taking CS into
1990; Parker et al., 1995). The sociometric categories account increases the sensitivity of sociometric
established by Coie and colleagues (Coie & Dodge, measurement.

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
1388 DeRosier and Thomas

This study also extended beyond traditional CS Level:

sociometric classification by subgrouping the un-
2 I-OLow
classified group according to the extreme socio- 3 Moderate

metric status group to which members were biased. IHigh

.N 2
Significant sociobehavioral differences were found
across the subgroups of unclassified children that
mirrored those for the extreme social status groups.
The difference between these patterns was in terms V 0

of magnitude, where the score for the status group

was more extreme than that for the corresponding
unclassified subgroup (e.g., for bullying, .73 for the Popular Rejected Controversial Neglected
rejected group vs. .33 for UR subgroup). Similar Social Status Group
gender differences for bullying were also found. A
Figure4. Peer victimization as a function of social status and
great deal of information is lost when the traditional
classification strength (CS). Letters are displayed for status groups
parsing of sociometric status groups is applied and in which there were significant differences across CS categories.
the unclassified group is dropped. Children who do Subgroups with different letters were significantly different from
not meet statistical cutoffs for inclusion into a status one another for that status group.
group should not simply be abandoned. A major
strength of the current approach is the ability to
include unclassified children in our studies of social 1.5
CS Level:
processes. In effect, greater understanding of the E]-Low
processes that may promote transitions into and out 1 DModerate
of extreme status groups can be gained by including
the unclassified group in longitudinal investigations
- IHigh
-s b a
of peer relations.
As expected, this study revealed significant . 0_
differences in social adjustment as a function of CS.
However, there was not a simple linear relation a -0.5

between CS and adjustment. Across the adjustment

areas, children within the high CS subgroup tended
Popular Rejected Controversial Neglected
to show exponential increases in adjustment prob-
Social Status Group
lems compared with the moderate and low sub-
groups. For example, the risk for victimization was Figure 5. Social withdrawal as a function of social status and
tremendously greater for highly rejected children classification strength (CS). Letters are displayed for status groups
(see Figure 4) compared with low or moderately in which there were significant differences across CS categories.
Subgroups with different letters were significantly different from
rejected children.
one another for that status group.
In addition, the pattern of relation for CS varied
depending on the status group and adjustment area
of interest. For each area of adjustment, only a subset in adjustment? The cutoffs for CS used in this study
of status groups showed differential patterns as a were based on statistical probabilities, not psycho-
function of CS. For example, in the prediction of logical phenomenon. There may be more optimal
bullying behaviors (see Figures 2-3), CS was salient cutoffs that would more fully capture crucial thresh-
only for the controversial and rejected status groups olds.
for whom higher CS was associated with significant In this study, the pattern of results, particularly
increases in these behavior problems. For with- those for the rejected and controversial groups,
drawn behavior (see Figure 5), however, it was the provides support for Moreno's (1934) belief that
rejected and neglected groups who showed signifi- both dimensions of acceptance and rejection are
cant increases in social withdrawal as a function of crucial for understanding social behavior. For the
higher CS. rejected group, low-level rejected children were not
Clearly, more research is needed to explore fully seen as bullies or relationally aggressive nor as
the underlying processes linking CS, social status, socially withdrawn and were reportedly only mod-
and adjustment. In particular, are there thresholds of estly victimized. When a large portion of the peer
social experience, positive or negative, that are group was repulsed within the context of little, if
qualitatively different and define significant shifts any, attraction to counteract that repulsion (i.e.,

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
 Strengthening Sociometric Prediction 1389

CS Level i seen as leaders within the peer group, with the

a DILow leadership of highly controversial children being
1 ab IModerate rated as commensurate with that of highly popular
m bb
children. Bullies tend to engage in proactive aggres-
a 0.5 b sion (i.e., calculated aggression designed to achieve a
self-interested goal, including the goal to dominate,
0 and not generally motivated by anger; Crick &
Dodge, 1996; Schwartz et al., 1998) toward
other children (Pellegrini, Bartini, & Brooks,
1999). This study suggests that controversial chil-
dren, especially highly controversial children,
Popular Rejected Controversial Neglected
may be particularly likely to engage in proactive
Social Status Group
aggression.
Figure 6. Leadership as a function of social status and classifica- Along the same lines, although both rejected and
tion strength (CS). Letters are displayed for status groups in which controversial children experience equally high num-
there were significant differences across CS categories. Subgroups bers of peers who dislike them, social withdrawal
with different letters were significantly different from one another and victimization were not present for controversial
for that status group.
children. Bullies do not typically experience peer
victimization unless they also lack prosocial skills
(Parkhurst & Asher, 1992). Controversial children
highly rejected), victimization by peers was extreme have been found to be socially adept (Newcomb,
along with elevated social withdrawal. Similarly, Bukowski, & Pattee, 1993; Volling, MacKinnon-
aggressive behavior problems were greatest for Lewis, Rabiner, & Baradaran, 1993), and findings
rejected children at the moderately and highly from this study suggest that they are social leaders,
rejected levels. able to manipulate successfully peers and social
It may be that when a high level of peer dislike is situations (e.g., organizing exclusion, rumor spread-
combined with no positive social support, the stage ing). Though controversial children have many
is set for becoming a scapegoat for victimization. peers who dislike them, they are able to maintain
Bullies may see that child as an easy target because the social support of many peers who like them. This
he or she clearly lacks potential defenders in the peer context of peer support may reinforce their bullying
group. Social withdrawal may increase in response behavior by minimizing the risk of becoming victims
to increased victimization as victims attempt to of bullying themselves. Given that past research has
escape or avoid the bullies (Olweus, 1993). Increases shown that controversial children are particularly
in aggressive behavior may reflect the tendency for disliked by rejected children (Duncan & Cohen,
victims of bullying to engage in reactive aggression 1995), it may be that controversial children function
(i.e., aggression based in anger and carried out in as the bullies in their relationships with rejected
retaliation for an actual or imagined offense), victims. The specific form of this dyadic relationship
including forms of revenge or retribution toward warrants further investigation.
bullies (Schwartz, Proctor, & Chien, 2001). Although It is important to note that, contrary to expecta-
these social dynamics are suggested by the pattern of tions, the pattern of results for CS did not vary by
findings from this study, further research is needed gender. Gender differences, both alone and in
to tease apart the direction of effect(s). In particular, interaction with status, were found for outcome
longitudinal studies are needed to investigate levels and these patterns were consistent with past
whether increasing strength of rejection results in research (e.g., Coie, Dodge, & Kupersmidt, 1990).
greater peer victimization and adjustment changes However, these gender differences were not further
(e.g., elevated reactive aggression and social with- qualified by CS. Thus, it appears that the magnitude
drawal) over time. of attraction and repulsion were equally influential
The controversial group showed a different on the sociobehavioral adjustment of males and
pattern. At all levels, controversial children were females. It would be informative for future research
seen as bullies and relationally aggressive. Higher to examine whether the presence of a reciprocated
levels of controversy among peers were associated friendship tempered the impact of CS for females
with significantly higher levels of these behavioral more so than males because of the greater impor-
problems, particularly relational aggression. How- tance placed on intimate friendships by females
ever, at the same time, controversial children were (Asher et al., 1996).

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
1390 DeRosier and Thomas

In sum, the strength of a child's classification into sociobehavioral problems were present for those
a sociometric status group was associated with who fell more extremely within the rejected, con-
significant differences in the patterns of adjustment, troversial, and neglected social status groups. Dif-
and our ability to predict social behavior was ferent patterns of maladjustment were present
enhanced by taking CS into consideration. However, depending on both the specific status group and
further research is needed to determine the true the CS into that status group. This greater differ-
potential of the CS construct. Based on this study, we entiation among groups could enhance our ability to
may expect children's risk for future adjustment direct interventions to those in greatest need. Better
problems, as well as the stability of status group targeting of interventions, in turn, should increase
assignment, to vary as a function of CS. Longitudinal the efficacy of treatment. In addition, if future
research is needed to determine whether higher CS longitudinal research supports unclassified biases
is related to greater stability of sociometric status as forerunners to extreme status classification,
and increased risk for later maladjustment. Long- efficacy could be increased by promoting early
itudinal research is also needed to investigate the intervention efforts before entrance into an extreme
relation between unclassified biases and transitions status group (Coie, 1990; Greenberg, Domitrovich, &
across status groups. Developmental research in- Bumbarger, 2001; Parker et al., 1995).
vestigating whether the relation between CS and
adjustment varies depending on the age of the
References
children would also inform our understanding of
this construct. For example, given that peers take on Asher, S. R., Parker, J. G., & Walker, D. L. (1996).
an increasingly meaningful and influential role as Distinguishing friendship from acceptance: Implications
children mature through the elementary school for intervention and assessment. In W. Bukowski & A.

years and adolescence (Furman & Buhrmester, Newcomb (Eds.), The company they keep: Friendship in
1992), it may be that CS affects adjustment to a childhood and adolescence (pp. 366-405). New York:
greater extent in older versus younger children. Cambridge University Press.
Bell-Dolan, D. J., & Wessler, A. E. (1994). Ethical admin-
In addition, this study was limited by its mono- istration of sociometric measure: Procedures in use and
method nature of measurement using peer nomina-
suggestions for improvement. Professional Psychology:
tions to measure both sociometric status and
Research and Practice, 25, 23-32.
behavioral adjustment. Future research using multi- Berry, D. A. (1996). Statistics: A Bayesian perspective. New
ple informants and multiple areas of adjustment are York: Duxbury Press.
needed to evaluate this new algorithm more fully. Buhrmester, D., & Furman, W. (1987). The development of
More research is needed to establish whether CS also companionship and intimacy. Child Development, 58,
1101-1113.
enhances the prediction of maladjustment according
to other sources, such as self-report or teacher report. Cillessen, A. H., & Bukowski, W. M. (2000). Conceptualiz-
Examination of other indexes of maladjustment, ing and measuring peer acceptance and rejection. In A.
H. Cillessen & W. M. Bukowski (Eds.), Recent advances in
including academic and internalizing problems, are
crucial. the measurement of acceptance and rejection in the peer
system (pp. 3-10). San Francisco: Jossey-Bass.
In addition to theoretical significance, this study Cillessen, A. H., Bukowski, W. M., & Haselager, G. J.
highlights multiple clinical implications for the (2000). Stability of sociometric categories. In A. H.
treatment of children with peer problems. Since the Cillessen & W. M. Bukowski (Eds.), Recent advances in the
inception of Coie et al.'s (1982) sociometric cate- measurement of acceptance and rejection in the peer system
gories, little applied research has been directed (pp. 75-93). San Francisco: Jossey-Bass.
toward the controversial group. In large part, there Coie, J. D. (1990). Toward a theory of peer rejection. In S. R.
has been an assumption that this group was not at Asher & J. D. Coie (Eds.), Peer rejection in childhood (pp.
elevated risk for negative outcomes. However, the 365-402). New York: Cambridge University Press.
findings from this study as well as those of other Coie, J. D., & Dodge, K. A. (1988). Multiple sources of data
on social behavior and social status in the school: A
recent investigations (e.g., Underwood, Kupersmidt,
cross-age comparison. Child Development, 59, 815-829.
& Coie, 1996,found controversial girls at greatest risk
Coie, J. D., Dodge, K. A., & Coppotelli, H. (1982).
for teen pregnancy) bring that assumption into Dimensions and types of social status: a cross-age
question. Elevated aggressive behavior found in this perspective. Developmental Psychology, 18, 557-570.
study underscores the need to direct interventions Coie, J. D., Dodge, K. A., & Kupersmidt, J. B. (1990). Peer
toward this group. group behavior and social status. In S. R. Asher & J. D.
In addition, the usefulness of CS for targeting Coie (Eds.), Peer rejection in childhood (pp. 3-16). New
interventions is potentially promising. Elevated York: Cambridge University Press.

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
 Strengthening Sociometric Prediction 1391

Crick, N. R., & Dodge, K. A. (1996). Social information- mental psychopathology: Vol. 2 Risk, disorder, and adaptation
processing mechanisms in reactive and proactive (pp. 96-161). New York: Wiley.
aggression. Child Development, 67, 993-1002. Parkhurst, J. T., & Asher, S. (1992). Peer rejection in middle
Crick, N. R., & Grotpeter, J. K. (1995). Relational aggres- school: Subgroup differences in behavior, loneliness,
sion, gender, and social-psychological adjustment. Child and interpersonal concerns. Developmental Psychology,
Development, 66, 710-722. 28, 231 - 241.
DeRosier, M. E., Kupersmidt, J. B., & Patterson, C. P. (1994). Pellegrini, A. D., Bartini, M., & Brooks, F. (1999). School
Children's academic and behavioral adjustment as a bullies, victims, and aggressive victims: Factors relating
function of the chronicity and proximity of peer to group affiliation and victimization in early adoles-
rejection. Child Development, 65, 1799-1813. cence. Journal of Educational Psychology, 91, 216-224.
Duncan, M. K., & Cohen, R. (1995). Liking within the peer Rutter, M. (1983). Stress, coping, and development: Some
group as a function of children's sociometric status and issues and questions. In N. Garmezy & M. Rutter (Eds.),
sex. Child Study Journal, 25, 265-287. Stress, coping, and development in children (pp. 1-41).
New York: McGraw.
Ericson, N. (2001). Addressing the problem of juvenile
bullying. In Office of Juvenile Justice and Delinquency Schwartz, D., Dodge, K. A., Coie, J. D., Hubbard, J. A.,
Cillesen, A. H., Lemerise, E. A., et al. (1998). Social-
Prevention Fact Sheet. Washington, DC: U.S. Department
of Justice. cognitive and behavioral correlates of aggression and
Furman, W., & Buhrmester, D. (1992). Age and sex victimization in boys' play groups. Journal of Abnormal
differences in perceptions of networks of personal Child Psychology, 26, 431-440.
Schwartz, D., Proctor, L. J., & Chien, D. H. (2001). The
relationships. Child Development, 63, 103-115.
Greenberg, M. T., Domitrovich, C., & Bumbarger, B. (2001). aggressive victim of bullying: Emotional and behavioral
dysregulation as a pathway to victimization by peers. In
The prevention of mental disorders in school-aged
children: Current state of the field. Prevention & J. Juvonen & S. Graham (Eds.), Peer harassment in school:
Treatment, 4, 1-67. The plight of the vulnerable and victimized (pp. 147-174).
New York: Guilford.
Han, J., & Kamber, M. (2000). Data preparation: Min-max
Terry, R. (2000). Recent advances in measurement theory
normalization. In J. Grey (Ed.), Data mining: Concepts and
and the use of sociometric techniques. In A. H. Cillessen
techniques (pp. 97-127). San Francisco: Morgan Kauf-
mann.
& W. M. Bukowski (Eds.), Recent advances in the
measurement of acceptance and rejection in the peer system
Hart, J. F. (1968). Computer approximations (a derivation of
(pp. 27-53). San Francisco: Jossey-Bass.
algorithm 5666). New York: Wiley.
Underwood, M. K., Kupersmidt, J. B., & Coie, J. D. (1996).
Kupersmidt, J. B., & DeRosier, M. E. (in press). The role of
Childhood peer sociometric status and aggression as
peer relations in the development of negative outcomes:
predictors of adolescent childbearing. Journal of Research
Explanatory processes. In K. A. Dodge & J. B. on Adolescence, 6, 201- 223.
Kupersmidt (Eds.), Peer rejection in childhood: Causes,
Volling, B. L., MacKinnon-Lewis, C., Rabiner, D., &
consequences, and treatment.
Baradaran, L. P. (1993). Children's social competence
McDougall, P., Hymel, S., Vaillancourt, T., & Mercer, L.
and sociometric status: Further exploration of aggres-
(2001). The consequences of childhood peer rejection. In
sion, social withdrawal, and peer rejection. Development
M. R. Leary (Ed.), Interpersonal rejection (pp. 213 -247).
and Psychopathology, 5, 459 -483.
London: Oxford University Press.
Moreno, J. L. (1933). Psychological and social organization
of groups in the community. Proceedings & Addresses.
American Association on Mental Deficiency, 38, 224-242.
Moreno, J. L. (1934). Who shall survive? A new approach to the Appendix
problem of human interrelations. Washington, DC: Ner-
vous and Mental Disease. Statistical Procedures for Calculating Status Group Bias
Newcomb, A. F., Bukowski, W. M., & Pattee, L. (1993). To illustrate the derivation of these formulas, we
Children's peer relations: A meta-analytic review of show how to derive an estimate for one of the
popular, rejected, neglected, controversial, and average popular status conditions, that is, ZLM >0. Another
sociometric status. Psychological Bulletin, 113, 99-128.
way to express this condition is that the child must
Olweus, D. (1993). Victimization by peers: Antecedents
have more than average, or mean, number of LM
and long-term consequences. In K. H. Rubin & J. B.
Asendorpf (Eds.), Social withdrawal, inhibition, and votes. Because our sample distribution of scores only
shyness in childhood (pp. 315-341). Hillsdale, NJ: approximates the true population, we can only
Erlbaum. provide qualified estimates of where the population
Parker, J. G., Rubin, K. H., Price, J. M., & DeRosier, M. E. mean lies. The central limit theorem states that the

(1995). Peer relationships, child development and distribution of sample scores approximates a normal
adjustment: A developmental psychopathology per- bell-shaped distribution, with our sample mean
spective. In D. Cicchetti & D. J. Cohen (Eds.), Develop- approximating the population mean. We can use

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms
1392 DeRosier and Thomas

this theorem to estimate likelihood of the population ZLM, that is, that the population mean 9ILM lies in an
mean being within a particular confidence interval. interval between negative infinity and a given
The methodology for this is to measure the area child's ZLM score. An algorithm that derives the
under the normal curve between the points that area under that curve between negative infinity and
bound the interval (see Berry, 1996, for detailed a given value is called the CDF (see Hart, 1968).
discussion of this methodology). Usually, we use a Thus, P(ZLM>O) is equivalent to the CDF(ZLM).
fixed confidence coefficient to discover the upper Generalizing, to calculate the probability that the
and lower bounds of an interval. But in our case, we population mean lies some threshold T below the
know the interval's boundaries and solve for the student's score, or P(Zx > T), is given by CDF(Zx- T).
confidence coefficient. The confidence coefficient (or Similarly, the P(Zx<T) is given by 1-CDF(T-Zx).
probability, written P(O)) that a given child's ZLM For more information regarding the software used to
score is greater than zero is equivalent to the implement these calculations and how to access this
probability that the population mean is less than software, contact the first author.

This content downloaded from 178.250.250.21 on Mon, 11 Jul 2016 05:16:32 UTC
All use subject to http://about.jstor.org/terms