Piers-Harris Children’s Self-Concept Scale, Second Edition

PIERS-HA

RRIS 2M

AN

UA

LEllen V. Piers, Ph.D

., and David S. H

erzberg, Ph.D.

W-388B

Piers-Harris 2Piers-Harris Children’s Self-Concept Scale,SECOND EDITION

Ellen V. Piers, Ph.D.David S. Herzberg, Ph.D.

MANUAL

Western Psychological Services • 12031 Wilshire Boulevard, Los Angeles, California 90025-1251Additional copies of this manual (W-388B) may be purchased from WPS.

Please contact us at 800-648-8857, Fax 310-478-7838, or www.wpspublish.com.

wps®

wps

Publishers Distributorswps®

Published by

The Piers-Harris Children’s Self-Concept Scale (Piers,1963) was originally developed in the early 1960s to providea brief, self-report instrument for the assessment of self-concept in children and adolescents. As defined by the scale’soriginal authors, self-concept is a relatively stable set of atti-tudes reflecting both description and evaluation of one’s ownbehavior and attributes. Since its introduction, the Piers-Harris has enjoyed widespread acceptance among cliniciansand researchers, as well as praise from reviewers. The instru-ment’s stature is reflected in more than 500 citations in pro-fessional journals and books in psychology, education, andthe health sciences. These numerous references highlight thePiers-Harris’s vital role in the expansion of knowledge aboutself-concept and its relationship to behavior.

The Piers-Harris Children’s Self-Concept Scale,Second Edition (Piers-Harris 2) represents the culminationof a careful revision process. The general goals of this pro-cess were to enhance the ease of use and psychometric foun-dation of the test, while preserving the many characteristicsof the instrument that have contributed to its success. Thesegoals have been realized in a set of specific improvements,including new nationwide normative data, an updated itemset, enhanced interpretive guidelines, and modernized com-puter assessment tools. Nevertheless, the Piers-Harris 2 re-tains the familiar response format, self-concept scales, andexcellent psychometric properties of the original edition.Thus, the revised test should be easily integrated into re-search projects and clinical assessments that used the origi-nal Piers-Harris.

General Description

The Piers-Harris 2 is a 60-item self-report question-naire, subtitled The Way I Feel About Myself. It is designedfor administration to children who are at least 7 years oldand have at least a second-grade reading ability. The mea-sure can be used with adolescents up to 18 years of age.

The Piers-Harris 2 items are statements that expresshow people may feel about themselves. Respondents areasked to indicate whether each statement applies to them bychoosing yes or no. Several methods of administration areavailable: the Piers-Harris 2 AutoScore™ Form (WPS

Product No. W-388A), which is completed by the child andscored manually by the test administrator; mail-in and fax-informs (WPS Product Nos. W-388C and W-388Z), which arecompleted by the child and submitted to WPS for computerscoring and report generation; a PC program (WPS ProductNo. W-388Y), which can generate a report based on eitheronline administration or offline data entry; and the SpanishAnswer Sheet (WPS Product No. W-388E), which is com-pleted by the child, whose answers are then transcribed ontoan AutoScore™ Form by the examiner. Using any of thesemethods of administration, most respondents can completethe Piers-Harris 2 in 10 to 15 minutes.

The Piers-Harris 2 includes the same Self-Conceptand Validity scales as the original Piers-Harris. The Self-Concept scales comprise the Piers-Harris 2 Total (TOT)score, which is a general measure of the respondent’s overallself-concept, and the six domain scales, which assess spe-cific components of self-concept. The domain scales includeBehavioral Adjustment (BEH), Intellectual and SchoolStatus (INT), Physical Appearance and Attributes (PHY),Freedom From Anxiety (FRE), Popularity (POP), andHappiness and Satisfaction (HAP). (On the original Piers-Harris, the Freedom From Anxiety scale was labeledAnxiety and the Behavioral Adjustment scale was labeledBehavior. All other scale names are unchanged from theoriginal instrument.) The Self-Concept scales are scored sothat a higher score indicates a more positive self-evaluationin the domain being measured. The Piers-Harris 2 Validityscales include the Inconsistent Responding (INC) index,which is designed to identify random response patterns, andthe Response Bias (RES) index, which measures a child’stendency to respond yes or no irrespective of item content.

Piers-Harris 2 Improvements

The most important feature of the Piers-Harris 2 is itsincorporation of new, nationally representative normativedata. The new norms are based on a sample of 1,387 students,aged 7 to 18 years, who were recruited from school districtsall across the United States. The sample closely approximatesthe ethnic composition of the U.S. population (U.S. Bureauof the Census, 2001a). The new standardization sample is a

1INTRODUCTION

3

significant improvement over the sample used to norm theoriginal Piers-Harris. That sample was recruited in the early1960s from a single public school system in rural Pennsyl -vania, and was relatively homogenous in terms of ethnicityand several other key demographic variables. In addition,whereas the original Piers-Harris sample consisted of 4ththrough 12th graders, the Piers-Harris 2 sample included2nd and 3rd graders as well.

The second major enhancement in the Piers-Harris 2is the reduction of the scale from 80 to 60 items. This itemreduction shortens administration time significantly, whileretaining all of the Self-Concept and Validity scales from theoriginal Piers-Harris. The deleted items included those ofrelatively less psychometric value, as well as those writtenin outdated language that was difficult for many children tounderstand. The revised scales are psychometrically equiva-lent to their counterparts in the original measure. Table 1summarizes the changes in item composition and labelingbetween the original and revised Self-Concept scales.

A third substantial change in the Piers-Harris 2 in-volves the microcomputer administration and scoring pro-gram. WPS offers a variety of computer services for many ofits products. The “Computerized Services for the Piers-Harris 2” section at the back of this manual provides infor-mation about the options available for the Piers-Harris 2.The software has been updated for the latest version of theMicrosoft Windows operating system, with an attractivenew graphical user interface. In addition, the computer re-port has been streamlined and updated to reflect the newnormative data.

This manual includes several new enhancements, in-cluding a revised section on interpreting the test that incor-porates three new case studies. Furthermore, the manualnow includes a topic-by-topic inventory of existing Piers-Harris studies (see Appendix A), to facilitate further re-search on the scale.

Principles of Use

The Piers-Harris 2 is appropriate for use in any re-search, educational, or clinical setting that requires efficient

quantitative assessment of children’s reported self-concept.The original Piers-Harris gained widespread acceptanceamong researchers, as reflected in an extensive scholarly lit-erature that has accumulated over the past four decades. Theinstrument has been used to evaluate psychological and edu-cational interventions, to investigate the relationship be-tween self-concept and other traits and behaviors (e.g.,empathy, teenage pregnancy, drug and alcohol use), and tomonitor changes in self-concept over time, among manyother research projects.

Because it is easily administered to groups, the Piers-Harris 2 can be employed as a screening device in class-rooms to identify children who might benefit from furtherpsychological evaluation. The Piers-Harris 2 can also beused in individual clinical assessments of children and ado-lescents. The Self-Concept scales can be used to generatehypotheses for clinical exploration, as well as to guide clin-icians in choosing among possible interventions and formu-lating referral questions for further psychological testing.

The Piers-Harris 2 can be administered and scored byteachers and other trained paraprofessionals. However, ulti-mate responsibility for its use and interpretation should beassumed by a professional with appropriate training in psy-chological assessment. Before administering the Piers-Harris 2, potential users should read this manual to becomefamiliar with the theoretical rationale, development, stan-dardization, and psychometric properties of the measure.

As with many self-report measures, users should keepin mind that the intent of the Piers-Harris 2 is readily appar-ent to most children and adolescents. For this reason, the re-sponses may be subject to conscious and unconsciousdistortion, usually in the direction of greater social desir-ability. The issue of response validity is addressed in greaterdetail in chapter 3 of this manual.

Although the Piers-Harris 2 is a useful instrument, itcannot by itself provide a comprehensive evaluation of achild’s self-concept. Such an evaluation is a complex taskrequiring clinical sensitivity and familiarity with the appli-cable research literature. In making clinical judgments con-cerning Piers-Harris 2 results, users should be prepared tointegrate other sources of data, which may include clinical

4

Table 1Self-Concept Scales of the Original Piers-Harris and the Piers-Harris 2

Original Piers-Harris Piers-Harris 2

Scale name No. of items Scale name No. of items

Total 80 Total (TOT) 60

Cluster scales Domain scales

Behavior (BEH) 16 Behavioral Adjustment (BEH) 14Intellectual and School Status (INT) 17 Intellectual and School Status (INT) 16Physical Appearance and Attributes (PHY) 13 Physical Appearance and Attributes (PHY) 11Anxiety (ANX) 14 Freedom From Anxiety (FRE) 14Popularity (POP) 12 Popularity (POP) 12Happiness and Satisfaction (HAP) 10 Happiness and Satisfaction (HAP) 10

Note. Some items are assigned to more than one scale. See Appendix E for a list of original items that were deleted in developing the Piers-Harris 2.

interviews with the child and other informants, prior history,school records, classroom observations, and results from otherpsychological tests. Users should also be prepared to conferwith outside consultants and referral sources as needed.

Contents of This Manual

Chapter 2 of this manual contains instructions for ad-ministering and scoring the Piers-Harris 2, and includes acompleted sample of an AutoScore™ Form. Chapter 3 pre-sents guidelines for interpreting the test results. Technicalaspects of the test are presented in chapters 4 and 5. Chapter4 reviews the development of the original Piers-Harris anddescribes the new standardization sample and item revisions

for the Piers-Harris 2. Chapter 5 discusses the reliability andvalidity of the Piers-Harris 2 and presents an overview of re-search on the technical properties of the original test. Thismanual also includes several appendixes that support spe-cialized applications of the test: Appendix A presents a listof research studies employing the Piers-Harris, organized bytopic; Appendix B reviews the use of the Piers-Harris withexceptional children; and Appendixes C and D contain in-structions and tables for comparing raw scores from thePiers-Harris 2 with those from the original version of thetest. Appendix E lists the items from the original Piers-Harris that were omitted from the Piers-Harris 2. Finally, inthe back of the manual is a chapter that provides instructionsfor using the Piers-Harris 2 computer-scoring products.

5

The original Piers-Harris Children’s Self-ConceptScale was developed in the 1960s as a research instrumentand as an aid for clinical and educational evaluation in ap-plied settings (Piers, 1984). Since its introduction, the Piers-Harris has functioned well in these roles, forming the basisfor an impressive and growing body of research. The Piers-Harris 2 is the first major revision and restandardization ofthe original Piers-Harris. The new features of the Piers-Harris 2, which include new normative data and an updateditem set, were implemented with the goal of maintaining asmuch backwards compatibility” as possible with the originalPiers-Harris. This chapter begins by reviewing the theoreticalrationale for the Piers-Harris and the development of theoriginal item set and scoring system. The remainder of thechapter is devoted to the Piers-Harris 2 revisions. The newstandardization sample is described first, followed by a dis-cussion of the item and scale changes. The chapter concludeswith a discussion of moderator variables and their effects oninterpretation. Users interested in making direct comparisonsbetween Piers-Harris 2 scores and scores on the originalmeasure should consult Appendixes C and D, which presentinstructions and tables necessary for comparing the scores.

Original Rationale andTheoretical Background

The original version of the Piers-Harris was based onthe view that individuals maintain relatively consistent be-liefs about themselves, beliefs that develop and stabilizeduring childhood. This set of beliefs represents a person’sself-concept, a term which some researchers have used interchangeably with terms such as self-esteem and self-regard. The original authors of the Piers-Harris assumedthat children would reveal important aspects of this underly-ing self-image by agreeing or disagreeing with simple,self-descriptive statements, and that this assessment of self-concept would relate meaningfully to other aspects of per-sonality and to predictions of future behavior.

From a global perspective, the term self-concept refersto a person’s self-perceptions in relation to important as-pects of life. Although shaped by biological and cultural fac-tors, these perceptions are formed primarily through the

interaction of the individual with the environment duringchildhood, and by the attitudes and behaviors of others.These perceptions give rise to self-evaluative attitudes andfeelings that have important organizing functions and thatalso motivate behavior. Over time, an individual’s self-concept may change in response to environmental or develop-mental changes, or as a result of changes in priorities orvalues. However, these changes usually do not occur rapidlyor in response to isolated experiences or interventions.

This definition of self-concept rests on several theo-retical assumptions:

1. Self-concept is essentially phenomenological innature. It is not something that can be observeddirectly but must be inferred from either behav-iors or self-report. Although behaviors aredirectly measurable, it is difficult to use behav-ioral observations to draw inferences about self-concept that are replicable and consistent acrossdifferent situations. Self-report, although subjectto many types of distortions, is closer to the pre-sent definition of self-concept, because it is adirect expression of the individual’s experienceof the self. The problem of distortion of self-report is a methodological issue, not a theoreti-cal one.

2. Self-concept has both global and specific com-ponents. Global self-concept reflects how anindividual feels about all the characteristics thatmake up his or her person, taking into account,among other things, skills and abilities, interac-tions with others, and physical self-image.Various specific aspects of self-concept resultfrom an individual’s self-appraisal in particularareas of functioning. These specific facets ofself-concept differ on several dimensions. Someare relatively broad (e.g., physical self, moraland ethical self, academic self); others are nar-rowly defined (e.g., good at mathematics, notskilled at baseball). The relative significance tothe individual of each aspect of self-concept de -termines the degree to which success and failure

4DEVELOPMENT AND RESTANDARDIZATION

37

affect overall self-evaluation (Dickstein, 1977;Harter, 1978). In unimportant areas, for example,perceived failure is not likely to have a strongimpact on the individual’s global self-evaluation.Similar notions have been proposed byShavelson, Hubner, and Stanton (1976), who viewself-concept as being “hierarchically organized.”

3. Self-concept is relatively stable . Althoughshaped by experience, it does not change easilyor rapidly. In children, self-concept is initiallymore situationally dependent and becomesincreasingly stable over time. Although it maybe possible to enhance children’s self-conceptthrough lengthy corrective experiences, changesare not likely to occur as the result of a brief,single, or superficial intervention. For example,a weekend camping trip may make a child feelgood but is unlikely to bring about lastingchange in that child’s self-concept. In addition,certain areas of self-concept may be more diffi-cult to change than others, and some may beamenable to change only during certain “criticalperiods” (Erikson, 1950; Schonfeld, 1969).

4. Self-concept has an evaluative as well as adescriptive component. It represents an individ-ual’s accumulated judgments concerning theself. Some of these evaluations may reflect in -ter nalized judgments of others (e.g., values,norms, and notions of what constitutes sociallydesirable traits and behaviors). Others may beunique to the individual. Thus, in evaluatingreported self-concept it is important to considerboth nomathetic (between-person) and idio-graphic (within-person) sources of comparison.The issues to be addressed concern both howchildren compare themselves to their peers andhow they evaluate themselves against their owninternal standards.

5. Self-concept is experienced and expressed dif-ferently by children at various stages of develop-ment. During infancy, the focus is on differ- entiating self from others and on establishing areciprocal relationship with the primary care-taker or caretakers (Ainsworth, 1979; Mahler,Pine, & Bergman, 1975). During the preschoolyears, the child becomes more mobile, interactssocially with other children and adults, and be -gins to develop a sense of gender identity. Self-concept during this period is defined primarilyby the child’s experience in each of these areas,and by parental attitudes and behaviors. Theconcepts of school-age children expand toencompass a larger arena of daily interactions,especially in the areas of achievement and peerrelationships. With increasing age and experi-

ence, the child’s self-perceptions also becomeincreasingly differentiated as he or she strugglesto integrate disparate aspects of experience intoa unified conceptual framework (Fahey &Phillips, 1981). In adolescence, certain aspectsof self-concept may undergo rapid change ordifferentiation (e.g., moral and ethical self-image, physical self-concept), whereas othersdevelop in a continuous, stable way (Dusek &Flaherty, 1981). For a more detailed discussionof developmental issues relating to self-concept,see Harter (1983).

6. Self-concept serves to organize and motivatebehavior. A stable self-concept maintains a con-sistent image of a person’s typical reactionsacross different situations. This helps to reduceambiguity in new situations and structure behav-ior toward preexisting goals. Action is alsoguided by an individual’s judgment of whetheror not a particular behavior is consistent with hisor her self-image. Behaviors that are congruentwith one’s self-concept will tend to be favoredover incongruent behaviors. In this fashion,judgments concerning the relative success orfailure of particular actions, as well as the emo-tions (e.g., pride, joy, humiliation) related tothese outcomes, may serve an important moti-vating function.

Original Piers-Harris Development

Item DevelopmentThe original Piers-Harris items were derived from the

work of Jersild (1952), who asked children what they likedand disliked about themselves. These statements were thengrouped into the following categories: (a) physical charac-teristics and appearance; (b) clothing and grooming; (c)health and physical well-being; (d) home and family; (e) en-joyment of recreation; (f) ability in sports and play; (g) aca-demic performance and attitudes toward school; (h)intellectual abilities; (i) special talents (music, arts); (j) “JustMe, Myself”; and (k) personality characteristics, inner re-sources, and emotional tendencies.

An initial item set, consisting of 164 items, was writtento reflect these various aspects of children’s self-concept.The items were written as simple declarative statements(e.g., “I am a happy person”), with a yes/no response format.To reduce the possible effects of response biases, approxi-mately half of the items were negatively worded (e.g., “I be-have badly at home”) and half were worded in the directionof positive self-concept (e.g., “I have many friends”). Mostitems were written to avoid such problematic features asdouble-negative constructions and ambiguous qualifierssuch as many, often, or rarely. Finally, 12 “lie” scale itemswere included to assess the tendency to respond in a social-ly desirable manner. These items were intended to mea-sure children’s willingness to admit relatively common

38 Technical Guide

weaknesses (e.g., “I am always good” or “Sometimes I actsilly”). However, these “lie” scale items were later droppedwhen it was found that they did not contribute significantlyto the validity of the scale.

This preliminary pool of items was then administeredto a sample of 90 children from Grades 3, 4, and 5. Tominimize errors due to differences in reading ability, theitems were read aloud by the examiners while the childrenfollowed along in their test booklets. This pilot study es-tablished that the children understood the items, and thatthe inventory could be completed in approximately 30 to35 minutes.

The pilot study results were used to reduce the itempool. Items answered in one direction by less than 10% ormore than 90% of the respondents were inspected and, inmost cases, dropped. However, because the instrument wasdesigned to identify children with problems in self-concept,a few items such as “My parents love me” were temporarilyretained even though answered yes by the great majority ofchildren. This procedure reduced the item pool to 140 items.

A second pilot study was conducted with a sixth-gradesample of 127 students. The 30 highest and 30 lowest scoreswere identified, and items were retained only if (a) they dis-criminated significantly between these high and low groups(p < .05), and (b) they were answered in the expected direc-tion by at least half of the high-scoring group. These proce-dures reduced the item pool to the 80 items that composedthe original Piers-Harris.

Scale ConstructionTotal score. The Piers-Harris Total score was based

on all 80 items. It was calculated by crediting 1 point foreach item answered in the direction of positive self-concept.The Total score was designed to measure a general dimen-sion of self-concept or self-esteem.

Cluster scales. Piers (1963) investigated the multi-dimensional nature of the scale by conducting a principalcomponents analysis using a sample of 457 sixth graders.Varimax rotation yielded six interpretable factors that to-gether accounted for 42% of the variance in item responses.These factors were labeled Behavior, Intellectual and SchoolStatus, Physical Appearance and Attributes, Anxiety,Popularity, and Happiness and Satisfaction.

This factor structure was supported by numerous sub-sequent factor analyses, which are described in more detailin chapter 5. The factor structure formed the basis for thesubscales of the Piers-Harris, which were called clusterscales in the 1984 edition of the Piers-Harris Manual). Aswith the Total score, the cluster scales were scored by cred-iting 1 point for each item answered in the direction of pos-itive self-concept.

Piers-Harris 2 Development

In revising the original Piers-Harris, the primary ob-jectives were to update and improve the test’s normative dataand item set. The original Piers-Harris normative data were

problematic in several respects. First of all, the standard-ization sample was recruited in the early 1960s from a singlepublic-school system in rural Pennsylvania. The standard-ization sample was relatively homogenous with respect toethnicity and several other key demographic variables. Thismade it more difficult to interpret Piers-Harris results for mi-nority children and those from other groups who differedsubstantially from the children in the standardization sample.Secondly, when the cluster and validity scales were devel-oped for the original Piers-Harris, they were normed on dif-ferent samples than the standardization sample used to normthe Total score. Although this procedure was necessary inorder to develop these important new scales, it clearly deviat-ed from ideal test-standardization procedures, in which allscales are normed on a single standardization sample.

A second set of concerns addressed in the revisionprocess involved the test items themselves. Decades of ex-perience with the original scale have revealed problems withcertain items. Some of these items have become difficult foryounger children to understand because they were writtenwith outdated language or low-frequency words. In addition,several researchers have identified opportunities to shortenthe Piers-Harris by deleting items that have relatively limit-ed psychometric utility.

Standardization SampleThe Piers-Harris 2 restandardization was based on a

large sample of students recruited from elementary, middle,junior high, and high schools throughout the United States.Table 6 presents the demographic characteristics of the sam-ple, along with corresponding percentages from the U.S.Census (U.S. Bureau of the Census, 2001a, 2001b) for com-parison. The sample is distributed relatively uniformly with-in the age range of 7 to 18 years (Grades 2 through 12). Thedistribution of the sample among ethnic groups is similar tothe U.S. Census figures, with slight underrepresentation ofAsians and Hispanics. Geographical distribution is ade-quate, with slight underrepresentation of participants fromthe western region of the United States.

Table 6 includes the distribution of a subsample ofparticipants among various categories of head-of-householdeducational level. This variable is used as an index of socio-economic status (SES), with higher educational level as-sociated with higher SES. The subsample is fairly close tothe U.S. Census figures for the lowest two SES categories.The proportion of subsample participants in the top SES cat-egory is higher than the U.S. Census proportion. The use ofa subsample for the SES comparisons reflects the fact thateducation data for heads of household were available foronly 673 participants in the Piers-Harris 2 standardizationstudy. The remaining 714 participants came from siteswhere SES data were not collected as part of the Piers-Harris 2 standardization study. Fortunately, these sites wereconducting other standardization studies concurrently, andSES data were collected in these other studies. These otherstandardization studies used different participants than thePiers-Harris 2 study. However, because sampling for the

Chapter 4 Development and Restandardization 39

various studies was random, there was no reason to expectsystematic differences in SES between participants in thePiers-Harris 2 study and those in other standardization stud-ies. Examination of the head-of-household education datafor participants in these other studies revealed a distributionthat is almost identical to the distribution for the 673 Piers-Harris 2 participants (see Table 6). Therefore, it is reason-able to assume that the distribution of head-of-householddata presented in Table 6 approximates that of the entirePiers-Harris 2 sample.

Item and Scale RevisionsItem revisions. As noted above, a major goal of the

Piers-Harris 2 revision was to streamline the scales by elim-inating problematic items. The first items targeted were thosethat contributed only to the Piers-Harris Total score and notto any of the six domain scores (Benson & Rentsch, 1988). Asecond set of items tagged for elimination included thosewhose wording has become outdated (e.g., “I have lots ofpep”); those that seemed specific to one sex (e.g., “I have agood figure”); and those containing words that frequentlyneeded additional explanation, especially for younger chil-dren (e.g., “I am obedient at home”). These procedures iden-tified 20 items as candidates for deletion, which left a revisedset of 60 items. Statistical analyses (which are described inthe next two sections) established that deletion of these itemswould result in no appreciable loss of reliability in the Totalscore or domain scale scores. In addition, the Total and do-main scores from the 60-item set correlated very highly withthe original scores derived from the 80-item set.

These analyses supported the decision to shorten themeasure by deleting the 20 candidate items. The deleteditems, with their original item numbers, are listed in Appen -dix E. (The remaining items, which constitute the Piers-Harris 2, are presented in Table 4.) In addition to thedeletions, one item was slightly altered from its originalwording. Item 37 was changed from “I am among the last tobe chosen for games” to “I am among the last to be chosenfor games and sports.” This change was made to ensure par-allel wording with two other items (9 and 51) that containthe phrase “games and sports.” The changes resulted in the60-item Piers-Harris 2, which yields the same scores as theoriginal test without sacrificing its psychometric strengths.The new version also decreases administration time signifi-cantly, as compared to the original measure. A readabilityanalysis was conducted on the Piers-Harris 2 item set. TheFlesch Reading Ease score was 91.8, and the Flesch-KincaidGrade Level was 2.1, indicating that second-grade readersshould be able to read the items with little difficulty.

Total (TOT) score. As in the original measure, ThePiers-Harris 2 Total (TOT) score is a measure of generalself-concept or self-esteem. The TOT score is derived bycrediting 1 point for each item answered in the direction ofpositive self-concept. Because all 80 items of the originalPiers-Harris were administered to the Piers-Harris 2 standard-ization sample, the reliability of the original and revisedscales could be compared (see chapter 5 for a more thoroughdiscussion of reliability and associated terminology).Coefficient alpha values for the 80-item original Total scoreand the 60-item Piers-Harris 2 TOT score were .93 and .91,respectively. These statistics indicate that both versionsdemonstrate robust internal consistency and that the revisedscale shows no significant loss in reliability compared to thelengthier original version. In addition, the original and re-vised Total scores correlate at .98, indicating that they arefunctionally equivalent.

Domain scales. As noted earlier, the original Piers-Harris included six factor-analytically derived cluster scales

40 Technical Guide

Table 6Demographic Characteristics of thePiers-Harris 2 Standardization Sample

U.S.Sample Census

n % %a

SexMale 689 49.7 51.3Female 698 50.3 48.7

Age in years 7–8 188 13.69–10 231 16.711–12 277 20.013–14 271 19.515–16 255 18.417–18 165 11.9

Race/Ethnic background Asian 17 1.2 3.5Black 255 18.4 14.7Hispanic/Latino 102 7.4 17.1White 943 68.0 60.9Native American 16 1.2 0.9Other 50 3.6 2.9Not Specified 4 0.2

U.S. Geographic RegionNortheast 316 22.8 19.0Midwest 424 30.6 22.9South 463 33.4 35.6West 184 13.3 22.5

Head of household’seducational levelb

Less than high school graduate 73 10.8 11.4High school graduate 182 27.0 31.9Some college 100 14.9 28.0Four-year college degree 318 47.3 28.7or more

Note. N = 1,387 (except for head of household’s educational level).aU.S. Census figures for sex, age, race/ethnicity, and geographic region(U.S. Census Bureau, 2001a) are based on U.S. population of school-aged children. Census data for head of household’s education level(U.S. Census Bureau, 2001b) is based on adults aged 25 to 54 (thosemost likely to be parents of school-aged children). bN = 673; see textfor discussion of missing data in this subtable.

(Behavior, Intellectual and School Status, PhysicalAppearance and Attributes, Anxiety, Popularity, and Happi -ness and Satisfaction). These scales have been retained inthe Piers-Harris 2, with very slight modifications. They arescored by crediting 1 point for each scale item answered inthe direction of positive self-concept.

Several labeling changes have been instituted in thePiers-Harris 2. The cluster scales have been relabeled do-main scales to reflect the fact that they are intended to mea-sure particular domains of self-concept. This label wasjudged more appropriate than cluster scales, which refers tothe statistical procedures used to assign items to the scales,rather than the clinical utility of the scales themselves. In ad-dition, the Anxiety scale is labeled Freedom From Anxiety inthe Piers-Harris 2. This was done to correct a confusing as-pect in the scoring of the original instrument. In the originalPiers-Harris, as in the Piers-Harris 2, all scales are scored sothat higher scores reflect more positive self-concept. On theoriginal instrument, the scale labeled Anxiety creates confu-sion because most users intuitively believe that higher scoresindicate more anxiety, but in fact higher scores on this scaleindicate less anxiety. The Piers-Harris 2 name FreedomFrom Anxiety removes this source of confusion and is moreconsistent with the labeling conventions for the other fivedomain scales, which generally refer to positive attributes(e.g., Popularity, Happiness and Satisfaction). Finally, theBehavior scale has been given the more specific, descriptivelabel of Behavioral Adjustment in the Piers-Harris 2.

The item assignments for the Piers-Harris 2 domainscales (see Table 3) represent relatively minor changes fromthe composition of the original cluster scales. The item re-visions did not affect the Freedom From Anxiety (FRE),Popularity (POP), and Happiness and Satisfaction (HAP)scales, which retain the same items as their counterparts inthe original measure. The Behavioral Adjustment (BEH)and Physical Appearance and Attributes (PHY) scales eachlost two items in the revision, and the Intellectual andSchool Status (INT) scale lost one item. Most of the 20items dropped for the Piers-Harris 2 contributed to only theTotal and not to any of the domain scales. Table 7 presentsreliability coefficients for the revised and original versionsof these three scales, demonstrating that the item changesdid not cause a meaningful decrement in the internal consis-tency of these scales. In addition, the extremely high corre-

lation coefficients demonstrate that the revised and originalversions of these scales are essentially equivalent.

As Table 3 indicates, the Piers-Harris 2 domain scales,like their counterparts in the original measure, contain nu-merous overlapping items. The item overlap is an artifact ofthe factor-analytic procedures used to derive the originalscales. Cooley and Ayres (1988) advocated eliminating thisitem overlap in order to increase the independence of thecluster scales. They suggested assigning each overlappingitem only to the scale with which it correlated most strongly.This procedure was considered for the Piers-Harris 2, but re-liability analyses showed that nonoverlapping domain scalessuffered a significant drop in internal consistency. This lossof reliability was especially apparent in the youngest chil-dren in the Piers-Harris 2 standardization sample. For 7- to8-year-old children, coefficient alpha was less than .60 forone scale and less than .70 for two others. Because such lowreliability coefficients are undesirable, the authors of thePiers-Harris 2 decided against the creation of nonover-lapping domain scales.

Validity scales. The Piers-Harris 2 includes twoValidity scales, the Inconsistent Responding (INC) indexand the Response Bias (RES) index, which help the test userto detect deviant response sets. The INC scale is designed toidentify random response patterns. It is based on the suppo-sition that certain pairs of responses are contradictory and/orstatistically improbable. The INC scale was introduced inthe 1984 edition of the Piers-Harris Manual, but has beenrevised extensively for the Piers-Harris 2.

Inconsistent Responding index. The INC scale wasconstructed using both rational and empirical procedures.First, the 60 Piers-Harris 2 items were examined to deter-mine pairs of items on which it is possible to produce logi-cally inconsistent responses (e.g., Item 3, “It is hard for meto make friends,” and Item 41, “I have many friends”).Second, a correlation matrix was formed for all 60 itemsusing the Piers-Harris 2 standardization data. Item pairs thatcorrelated at r ≥ .25 were examined, and pairs were retainedif their content created the potential for a logically inconsis-tent pair of responses, as in the aforementioned example.Third, frequency tables were constructed for all item pairsidentified in the first two steps. These tables were used todetermine which particular combination of responses (e.g.,yes on one item, no on another) occurred least frequently in

Chapter 4 Development and Restandardization 41

Table 7Intercorrelations and Reliability Coefficients for Revised and Original Domain Scales

Coefficient alpha

Domain scale r a Piers-Harris 2 Original Piers-Harris

Behavioral Adjustment (BEH) .98 .81 .81Intellectual and School Status (INT) .99 .81 .82Physical Appearance and Attributes (PHY) .98 .75 .79

Note. These are the only domain scales (called “cluster scales” on the original Piers-Harris) that had item changes in the Piers-Harris 2 revision. aThe correlation coefficient between the revised and original scale raw scores.

the standardization sample. Item pairs were retained if theyproduced a pair of logically inconsistent responses that oc-curred in less than 10% of the sample.

These procedures resulted in an INC scale composedof 15 item pairs, which are displayed in Table 8. This revisedINC scale features two substantial improvements over theone described in the 1984 edition of the Piers-HarrisManual. First of all, the current revision includes no dupli-cate items among its 15 item pairs, whereas the 25 item pairsin the original version of the INC scale included numerousduplicate items. This meant that a particular response to asingle item could have a disproportionate effect on the totalINC score, potentially distorting the meaning of the scale.Secondly, the items in the current revision of the scale aredistributed relatively evenly among the six domain scales.The original version of the INC scale was heavily weightedwith items from the Behavior scale, with fewer items repre-senting the remaining five domain scales. This meant thatthe original INC score gave little information about the va-lidity of responses on these other five scales.

The INC index is scored by crediting 1 point for eachpair of items for which the examinee gives the specifiedcombination of keyed responses. For example, a yes re-sponse to Item 1 (“My classmates make fun of me”) coupledwith a no response to Item 47 (“People pick on me”) wouldbe scored “1” on the INC index. Note that the other combi-nation of logically inconsistent responses to this item pair(no to Item 1, yes to Item 47) would not be scored on theINC index. This scoring system results in a possible INCindex score of 0 to 15.

In the Piers-Harris 2 standardization sample, the INCscore had a mean of .89 and a standard deviation of 1.12. Aswould be expected, the distribution of this score was highlypositively skewed (that is, most examinees produced verylow scores on the INC index). The cumulative frequenciesfor the INC index are shown in Table 9.

As noted in chapter 3, interpretation of the INC indexis based on a cutoff score of 4. When a child scores 4 orgreater on this index, there is a very high likelihood that the

42 Technical Guide

Table 8Inconsistent Responding (INC) Index Item Pairs

Frequencyof responsecombination

Inter- in standard-item ization

INC item pair (Keyed response) correlation sample (%)

1. My classmates make fun of me. (Y) 47. People pick on me. (N) .46 7.2

2. I am a happy person. (N) 42. I am cheerful. (Y) .45 4.5

3. It is hard for me to make friends. (N) 41. I have many friends. (N) .40 7.9

4. I am often sad. (N) 40. I am unhappy. (Y) .46 5.3

5. I am smart. (N) 43. I am dumb about most things. (N) .42 4.9

7. I get nervous when the teacher calls on me. (Y) 10. I get worried when we have tests in school. (N) .35 9.6

9. I am a leader in games and sports. (Y) 51. In games and sports, I watch instead of play. (Y) .27 5.3

14. I cause trouble to my family. (N) 20. I behave badly at home. (Y) .41 6.3

18. I am good in my schoolwork. (N) 21. I am slow in finishing my schoolwork. (N) .32 8.1

19. I do many bad things. (Y) 27. I often get into trouble. (N) .47 6.6

26. My friends like my ideas. (N) 39. My classmates in school think I have good ideas. (Y) .59 3.7

29. I worry a lot. (N) 56. I am often afraid. (Y) .36 6.0

31. I like being the way I am. (N) 35. I wish I were different. (N) .48 3.9

44. I am good-looking. (Y) 49. I have a pleasant face. (N) .56 6.5

53. I am easy to get along with. (Y) 60. I am a good person. (N) .38 3.3

Table 9Cumulative Frequency of

Inconsistent Responding (INC)Index Scores in the

Standardization Sample

INC score Cumulative frequency (%)

0 47.9

1 77.0

2 91.0

3 96.7

4 99.0

5 99.8

6 99.9

7 99.9

8 99.9

9 100.0

Note. N = 1,387.

Piers-Harris 2 was completed in a random or inconsistentmanner. This cutoff score was determined by first selecting ascore that was approximately 2 standard deviation unitsabove the mean. In the Piers-Harris 2 standardization sam-ple, the corresponding score was 3.13 (or 3, when roundedto the nearest whole number). As Table 9 illustrates, only3.3% of the children in the Piers-Harris 2 standardizationsample produced an INC index score greater than 3. Thissuggested that a cutoff score of 4 would identify those chil-dren whose INC scores were extremely deviant.

To further assess the interpretive value of this cutoffscore, analyses were undertaken to determine the probabili-ty that an INC score of 4 represented a random response set.To accomplish this, the distribution of INC scores from thePiers-Harris 2 standardization sample (N = 1,387) was com-pared to the distribution of INC scores from 1,387 randomPiers-Harris 2 response sets. The question of interest was:

For any given INC score, what was the probability that itwas drawn from the random data set as opposed to the datacollected from actual respondents? For each INC score,the ratio of the frequency counts between the two sampleswas calculated and expressed as a percentage indicatingthe probability that the score was from the random data.Table 10 presents these percentages for all possible scoreson the INC scale. Of the 337 cases that scored 4 on the INCscale, 32 were in the respondent sample and 305 were inthe random sample. Thus, any Piers-Harris 2 response setthat yielded an INC score of 4 had a 91% likelihood ofbeing a random response set.

These analyses suggest that a cutoff score of 4 is use-ful for interpreting this index. An INC score of 4 representsa high probability (91%) that a particular Piers-Harris 2 wascompleted at random, or without adequate understanding ofthe content of the test items.

Response Bias (RES) index. The RES scale is astraightforward measure of response bias. A positive responsebias represents the tendency to answer yes regardless of itemcontent, whereas a negative response bias represents the ten-dency to answer no regardless of item content. Either of thesebiases might distort the meaning of Piers-Harris 2 Self-Concept scores. The RES index allows the examiner to takeresponse bias into account when interpreting the Piers-Harris2. This index is a count of the number of items to which the ex-aminee has responded yes. The RES score has a possible rangeof 0 to 60. Very high RES scores indicate a positive responsebias, and very low scores signal a negative response bias.

In the Piers-Harris 2 standardization sample, the RESscore had a mean of 28.39 and a standard deviation of 5.28.The distribution of the scores approximated a normal distri-bution. Cutoff scores for interpretation were set at +/– 2standard deviation units. These cutoffs corresponded to RESscores of 40 and 18, respectively. Thus, children scoring 40or above on the RES scale are identified as having given an

Chapter 4 Development and Restandardization 43

Table 10Probability That Various

Inconsistent Responding (INC) ScoresRepresent Random Responding

Probability of randomINC score response (%)

0 2

1 21

2 52

3 81

4 91

5 96

6 99

≥7 >99

Table 11Descriptive Statistics for Piers-Harris 2 Raw Scores in the Standardization Sample

Scale No. of items M SD

Self-Concept Scales

Total (TOT) 60 44.6 10.2

Domain Scales

Behavioral Adjustment (BEH) 14 11.2 2.9

Intellectual and School Status (INT) 16 11.9 3.4

Physical Appearance and Attributes (PHY) 11 7.8 2.6

Freedom From Anxiety (FRE) 14 10.0 3.4

Popularity (POP) 12 8.4 2.7

Happiness and Satisfaction (HAP) 10 8.1 2.2

Validity Scales

Inconsistent Responding (INC) 15 pairs 0.9 1.1

Response Bias (RES) 60 28.4 5.3

Note. N = 1,387.

unusually large number of yes responses. Conversely, chil-dren scoring 18 or below on the RES index are flagged ashaving given an unusually large number of no responses.These two cutoff scores were chosen because they are reason-ably conservative and because moderate deviations in re-sponse set in either direction usually do not create majorproblems for interpretation.

Derivation of Standard ScoresTable 11 presents means and standard deviations for the

raw scores of the Piers-Harris 2 scales in the standardizationsample. To facilitate interpretation of Piers-Harris 2 results,these raw scores have been converted to normalized T-scores(see Anastasi, 1988, p. 88). The original distribution ofPiers-Harris 2 raw scores underwent a nonlinear transform-ation so that it would approximately fit a normal curve. Thenormalized raw scores were converted to T-scores, whichhave a mean of 50 and a standard deviation of 10. Becausethe Piers-Harris 2 uses normalized T-scores, every scale hasan approximately normal distribution of standard scores.Thus, the frequency of cases in the standardization samplefalling above or below a given T-score is nearly identical forevery scale. This simplifies intra-individual comparisonsamong the domain scales of the Piers-Harris 2. For exam-ple, if a child achieves normalized T-scores of 60 on theBEH and POP domain scales, the examiner knows that inthose areas of self-concept, the child has scored higher thanapproximately 84% of the children in the standardizationsample. The use of normalized T-scores also enables the ex-aminer to compare a particular child’s test results with thoseof the children in the standardization sample.

Moderator VariablesModerator variables are respondent characteristics, such

as age, sex, or ethnicity, that may affect scores on a psycho-logical test independently of the construct that the test isattempting to measure. When groups that differ on moderatorvariables also perform differently on a test, it can be difficultto interpret test results without separate norms for each group.This section discusses moderator variables that may impactPiers-Harris 2 results. The section also presents statisticalcomparisons that determine whether there are any meaning-ful differences between groups defined by the moderator variables in the standardization sample. One of the advantagesof the Piers-Harris 2 standardization sample, as comparedwith the original Piers-Harris standardization sample (Piers,1984), is its demographic representativeness. Whereas theoriginal sample was recruited from a single school district in Pennsylvania, the new sample consists of an ethnically diverse group of children from all regions of the UnitedStates. This allows statistical analysis of the potentialmoderating effects of ethnicity on Piers-Harris 2 scores.

With a sample as large as the Piers-Harris 2 standard-ization sample, it is necessary to set clear guidelines as towhat constitutes a practically meaningful difference betweenaverage group scores. Because the power to detect groupdifferences is so high in large sample analyses, differences

as small as 1 or 2 T-score points are often statistically signif-icant. From a practical perspective, however, a 1- or 2-pointchange in a test score yields virtually no new informationabout the individual who took the test. In addition, suchsmall differences are usually within the standard error ofmeasurement of the test, meaning that the difference is notlikely to be reliable across administrations of the test.

An effect size metric can be used to evaluate whetherstatistically significant differences between large groups arealso practically meaningful (Cohen, 1992). The relevantstatistic is calculated by dividing the mean difference betweengroups on a score by the standard deviation of that score forthe two groups combined. The use of T-scores simplifiesthese calculations. Because the pooled standard deviation al-ways approximates 10 for any T-score comparisons, 1 effect-size unit equals about 10 T-score points. Effect sizes of 1 to 3T-score points are considered small and not clinicallymeaningful; effect sizes between 3T and 5T are consideredmoderate; and those greater than 5T are considered large.

Moderator effects for Piers-Harris 2 scale scores wereevaluated by calculating average T-scores for groups definedby the moderator variables (e.g., males and females, ethnicgroups). These average group scores were then compared tothe average T-score for the entire standardization sample(which is 50 by definition). For each moderator, the follow-ing decision rule was applied: If the deviation of a particularmoderator group from the overall standardization samplewas equivalent to a small effect size, the moderator was con-sidered meaningless in practical applications. In these cases,stratification of the normative data was not required. If, onthe other hand, the deviation constituted a moderate to largeeffect size, the pattern of T-score differences among themoderator groups was examined more closely to determinewhether it was consistent with other known characteristicsof the groups under consideration. If robust, replicable pat-terns of group differences were present, the use of separatenorms for each group was considered.

Age. Many psychological theories predict substantialchanges in self-concept between certain phases of develop-ment (e.g., between adolescence and adulthood). Certainmeasures of self-concept, especially those designed for bothadolescents and adults, have employed age-stratified norms(e.g., Tennessee Self-Concept Scale, Second Edition; Fitts& Warren, 1996). However, empirical research has not sup-ported an association between age and self-concept scoresfor individuals between the ages of 8 and 23 (see Wylie,1979, for a review). Table 12 presents average T-scores forage groupings in the Piers-Harris 2 standardization sample.Out of all the possible comparisons, only one domain scalescore (INT) differed from the sample mean by more than 3T-score points, and this occurred only for a single age group(7- to 8-year-olds). There was no evidence of a consistentpattern of clinically significant age-related differences inPiers-Harris 2 scores. Thus, there appeared to be no need forage-stratified norms.

Sex. Generally speaking, investigators have failed tofind significant sex differences in self-concept (Hattie, 1992;

44 Technical Guide

Wylie, 1979). Early studies using the Piers-Harris as a mea-sure of general self-concept (e.g., Piers & Harris, 1964) wereconsistent with this notion.

However, more recent studies have suggested sex dif-ferences in specific aspects of self-concept. In particular,males tend to report less anxiety and more problematic be-haviors than girls. Lewis and Knight (2000) examined moder-ator variables for the original Piers-Harris in 368 intellectuallygifted children in Grades 4 to 12. Males and females did notdiffer on the Total score, but there were sex differences onthree cluster scales. Females scored significantly higher thanmales on the Behavior and Intellectual and School Statusscales, whereas the reverse pattern occurred on the Anxietyscale. Similar sex differences on the Anxiety scale were re-ported by Lord (1971) and Osborne and LeGette (1982).Furthermore, in the normative sample for the cluster scalesdescribed in the 1984 edition of the Piers-Harris Manual,males rated themselves significantly lower in self-concept onthe Behavior scale than females. Additional studies related tosex differences are listed by topic in Appendix A.

Average T-scores for males and females in the Piers-Harris 2 standardization sample are provided in Table 13.There were slight sex differences on the BEH and FREscales, consistent with the previously cited literature. How -ever, the effect sizes were not large enough to constitute apractically meaningful difference. These findings do notsupport stratification of the Piers-Harris 2 norms by sex.

Ethnicity. The original Piers-Harris norms were basedon an ethnically homogenous sample of public-school chil-dren from one Pennsylvania school district (Piers, 1984).Because it was not possible to examine ethnic differences inself-concept using these data, the 1984 edition of the Piers-Harris Manual relied on an extensive review of the litera-ture documenting comparisons among various ethnic groupson Piers-Harris scores. Those studies, along with more re-cent literature relating to this issue, are listed by topic inAppendix A.

Certain themes emerge from this literature. Most im-portantly, ethnicity per se does not appear to be a significantdeterminant of self-concept. Rather, children in certain ethnicor cultural groups may be at higher risk for the kinds of on-going stressors that can impact self-esteem over the long run.These experiences may include racial discrimination, aca-demic failure due to language deficits or other problems, andgeneral difficulties adjusting to the mainstream culture. Onthe other hand, adequate social support seems to insulate chil-dren from at least some of these potentially harmful factors.

Cultural differences in response style must also betaken into account. In some cultures, saying positive thingsabout oneself is frowned upon because it is consideredboast ful. Also, individuals from certain cultures may be lessinclined to provide socially desirable responses on psycho-logical tests. Either of these factors could cause spuriouslylow Piers-Harris 2 scores. Alternatively, spuriously highscores may result in individuals from groups that tend to bemore defensive or reluctant in terms of disclosing negativeemotions.

Chapter 4 Development and Restandardization 45

Table 12Average T-Scores by Age Group in the Piers-Harris 2 Standardization Sample

Age

7–8 9–10 11 –12 13–14 15–16 17–18

Self-Concept scale (n = 188) (n = 231) (n = 277) (n = 271) (n = 255) (n = 165)

Total (TOT) 51.2 51.5 51.3 48.4 48.6 49.1

Behavioral Adjustment (BEH) 51.2 50.9 51.4 47.8 47.6 50.1

Intellectual and School Status (INT) 53.3 52.0 50.0 47.5 48.1 49.2

Physical Appearance and Attributes (PHY) 50.4 50.2 50.2 48.4 50.6 49.5

Freedom From Anxiety (FRE) 49.9 50.6 50.7 50.4 48.2 48.8

Popularity (POP) 47.7 50.3 51.2 50.5 50.0 48.6

Happiness and Satisfaction (HAP) 50.7 51.0 50.7 47.9 48.5 48.8

Note. N = 1,387.

Table 13Average T-Scores by Sex in the

Piers-Harris 2 Standardization Sample

Sex

Male Female

Self-Concept scale (n = 689) (n = 698)

Total (TOT) 49.6 50.4Behavioral Adjustment (BEH) 48.2 51.3Intellectual and School Status (INT) 48.7 51.0Physical Appearance and Attributes (PHY) 49.6 50.1Freedom From Anxiety (FRE) 51.6 48.1Popularity (POP) 49.6 50.2Happiness and Satisfaction (HAP) 49.1 50.0

Note. N = 1,387.

The preceding remarks can be viewed as a caveat toconsideration of ethnic differences in the Piers-Harris 2norms. Unlike the 1984 manual’s norms, the new norms arebased on an ethnically diverse sample, and so it is possibleto investigate directly the moderating effects of ethnicity.The Piers-Harris 2 standardization sample included suffi-ciently large numbers of Black, Hispanic, and White partic-ipants to permit evaluation for meaningful ethnicdifferences. Table 14 presents average T-scores for thesegroups. None of the individual scale scores differed by morethan 3 T-score points from the grand mean for the standard-ization sample. These small effect sizes suggest that Piers-Harris 2 scores can be interpreted without the need forseparate norms for these groups.

Table 14 also presents average T-scores for Asians,Native Americans, and those who reported their ethnicity as“Other.” Because of the small cell sizes, these results are notreliable enough to aid interpretation of Piers-Harris 2 scores.Nevertheless, the findings may be of interest to those plan-

ning research on self-concept with the Asian or NativeAmerican populations.

Socioeconomic status. Generally speaking, the stud-ies that have examined socioeconomic status (SES) and self-concept have not found significant relationships betweenthese constructs (Hattie, 1992; Wylie, 1979). There are a fewexceptions to this trend. For example, Osborne and LeGette(1982) studied a sample of 374 middle school students andfound that lower SES was associated with lower self-con-cept scores. As noted with respect to ethnicity, the relation-ships between SES and self-concept are likely to becomplex, in that SES may be a marker for other variables,such as level of stress within the family, that affect self-concept more directly.

As noted in Table 6, SES data are available for abouthalf of the Piers-Harris 2 standardization sample, but there isgood reason to believe that these data provide a good ap-proximation of the SES distribution in the entire sample.Table 15 shows average T-scores for groups defined by the

46 Technical Guide

Table 14Average T-Scores by Race/Ethnicity in the Piers-Harris 2 Standardization Sample

Race/Ethnic group

NativeBlack Hispanic White Asian American Other

Self-Concept scale (n = 255) (n = 102) (n = 943) (n = 17) (n = 16) (n = 50)

Total (TOT) 48.7 47.4 50.8 48.0 50.9 46.8

Behavioral Adjustment (BEH) 48.4 47.7 50.5 48.7 50.6 47.1

Intellectual and School Status (INT) 48.9 47.6 50.6 47.9 50.4 46.8

Physical Appearance and Attributes (PHY) 50.7 48.8 49.8 48.8 49.2 49.0

Freedom From Anxiety (FRE) 48.1 47.5 50.8 46.6 51.1 47.0

Popularity (POP) 48.9 47.9 50.6 49.0 48.2 47.6

Happiness and Satisfaction (HAP) 48.0 48.4 50.2 51.4 51.2 47.5

Note. N = 1,383. Due to the small cell sizes for the Asian, Native American, and Other samples, the results for those groups are not reliable enough toaid in interpreting the Piers-Harris 2 score.

Table 15Average T-Scores by Head of Household’s Education Level in the Piers-Harris 2 Standardization Sample

Head of household’s education level

Not HS HS Some College Post-graduate graduate college graduate graduate

Self-Concept scale (n = 73) (n = 182) (n = 100) (n = 157) (n = 161)

Total (TOT) 47.0 49.8 50.8 51.9 50.3

Behavioral Adjustment (BEH) 46.6 48.2 49.6 51.3 50.0

Intellectual and School Status (INT) 47.2 49.2 50.8 51.5 50.1

Physical Appearance and Attributes (PHY) 48.5 50.7 50.8 50.8 50.9

Freedom From Anxiety (FRE) 47.2 50.5 49.1 51.6 50.1

Popularity (POP) 48.1 50.8 51.4 51.2 50.5

Happiness and Satisfaction (HAP) 47.3 49.6 49.8 51.3 49.9

Note. N = 673. See text for discussion of missing data.

reported education level of the head of household. There is atrend toward lower self-concept scores in the group that islowest on this SES index (the head of the household did notgraduate from high school), which is consistent with Osborneand LeGette’s (1982) findings. However, the differences be-tween this group’s scores and the grand means for the entirestandardization sample remain in the realm of small effectsizes (two scales, BEH and TOT, differ by slightly more than3 T-score points from the grand mean). Considering all ofthe SES groups, there is no consistent pattern of clinicallymeaningful score differences associated with increasingSES. These data do not indicate a need for Piers-Harris 2norms that are stratified by SES.

U.S. geographic region. Although there is no theoreti-cal reason to expect regional differences in Piers-Harrisscores, the possibility of such differences was explored in thecurrent data set. Table 16 presents the average T-scores byU.S. geographic region for the Piers-Harris 2 standardizationsample. As with the other moderators, there is no consistentpattern of clinically meaningful differences among regions.

Intelligence and academic achievement. Numerousinvestigators have examined the relationships among intelli-gence, academic achievement, and self-concept (see Ap -pendix A). Generally speaking, researchers have foundmoderate positive correlations between measures of achieve -ment and self-concept scores. Furthermore, as hypothesizedby Shavelson et al. (1976), the relationship between these twoconstructs appears to be due to a specific academic compo-nent of self-concept, rather than to generalized self-concept.For example, two studies of elementary-school and middle-school students demonstrated that specific measures of aca-demic self-concept were stronger predictors of achievementthan the Piers-Harris Total score (Lyon & MacDonald, 1990;Schike & Fagan, 1994). In contrast to achievement, re-searchers have typically found only weak associations be-tween intelligence test scores and measures of self-concept(e.g., Black, 1974; McIntire & Drummond, 1977).

With regard to the Piers-Harris 2 normative sample,the literature suggests that children who perform well on

academic tasks are likely to have higher than average scoreson the INT cluster scale, but not on the other domain scalesor the Total score. However, this must remain a tentativehypothesis. Academic achievement data is not available forthe Piers-Harris 2 standardization sample, so it is not possi-ble to study empirically the effects of this moderator.

Summary. The Piers-Harris 2 standardization samplehas been examined for group differences related to thepotential moderating variables of age, sex, ethnicity, socio-economic status (as indexed by education level of head ofhousehold), and U.S. geographic region. Where effects ofthese moderators were present, they were small and appliedonly to one or two domain scales. None of the analysesidentified clinically meaningful patterns of differences thatwere consistent with other knowledge about the groups inquestion. Con se quently, it was determined that one set ofnonstratified normative data could be used for interpretingPiers-Harris 2 scores.

Nevertheless, the relationship among moderators iscomplex and deserves additional study. Although the currentanalyses do not support stratified norms for the Piers-Harris2, several prior studies have found group differences on eachof the moderators described in the previous section. Thesestudies vary greatly in their methodological quality. The au-thors of the Piers-Harris 2 recognize that particular clinicaland research applications of this measure may raise con-cerns about the suitability of the nonstratified norms pre-sented in this manual. For example, special care should betaken in interpreting Piers-Harris 2 results for children withmental retardation, children with severe psychiatric dis-orders, and children from ethnic groups not well representedin the standardization sample. It is suggested that in theseand similar situations users familiarize themselves with theappropriate research literature to determine what caveatsshould be considered in using the Piers-Harris 2 norms. Tofacilitate this process, Appendix B of this manual includes abrief review of research relating to the use of the Piers-Harris for children with special needs. Other studies involv-ing special populations are listed by topic in Appendix A.

Chapter 4 Development and Restandardization 47

Table 16Average T-Scores by U.S. Geographic Region in the Piers-Harris 2 Standardization Sample

U.S. geographic region

Northeast Midwest South West

Self-Concept scale (n = 316) (n = 424) (n = 463) (n = 184)

Total (TOT) 50.8 50.8 49.4 48.2

Behavioral Adjustment (BEH) 50.2 50.9 48.3 49.9

Intellectual and School Status (INT) 50.5 51.3 48.8 47.9

Physical Appearance and Attributes (PHY) 51.0 49.9 50.3 46.8

Freedom From Anxiety (FRE) 49.9 50.4 49.9 48.4

Popularity (POP) 50.4 49.4 50.7 48.4

Happiness and Satisfaction (HAP) 50.4 50.2 48.6 49.2

Note. N = 1,387.

During the nearly four decades since the introductionof the Piers-Harris, numerous investigators have studied thetechnical characteristics of the instrument. This chapter reviews selected studies from that literature and presents newreliability and validity evidence from the Piers-Harris 2 standardization sample. As demonstrated in the previouschapter, the Piers-Harris 2 is essentially identical to the origi-nal measure from a psychometric perspective. Researchersmay therefore proceed to use the Piers-Harris 2 with the confidence that its reliability and validity are strongly upheldby the extensive database pertaining to the original scale.

Reliability

Reliability concerns the stability of scores on a psycho-logical test. A reliable test should produce consistent scoresfor the same individual when he or she takes the test ondifferent occasions and under varying conditions of exam-ination. A reliable test should also yield scores that arerelatively free of measurement error, or variance in a scoredue to chance factors rather than true variance in the psycho-logical construct being assessed. Reliability estimates areexpressed in terms of correlation coefficients that vary from0 to 1, with higher figures indicating greater reliability.Reliability is considered the most basic psychometric prop-erty of a test, because the discussion of whether a test mea-sures what it is supposed to measure cannot even begin untilthe test’s reliability has been established. In other words, re-liability is a necessary, but not a sufficient, condition forvalidity. This section will consider two aspects of reliability:internal consistency and test-retest reliability. The sectionwill describe new reliability data collected for the Piers-Harris 2 revision, and will also review the extensive litera-ture of reliability studies on the original Piers-Harris.

Internal Consistency of the Original andRevised Instruments

The reliability of a psychological test is determined inlarge part by how well the test items sample the content do-main being assessed. A set of test items that performs well inthis regard has the quality of internal consistency. In an

internally consistent test, items tend to be highly inter-correlated, presumably because they are measuring the sameconstruct. The primary indexes of internal consistency arecoefficient alpha (Cronbach, 1988), or, if the items aredi chotomous, Kuder-Richardson Formula 20 (KR-20; Kuder& Richardson, 1937). These statistics measure the averageintercorrelations among items in a test and are thought to establish an upper limit for the reliability of a test. Anothermethod for assessing the consistency of content sampling in a test is split-half reliability, in which the test is split into equivalent halves for each individual, and the relevantstatistic is the Pearson correlation between the two halves.

Piers-Harris 2. Table 17 presents internal consistencyestimates for the six domain scales and Total (TOT) scoreof the Piers-Harris 2. Alphas are reported for the entirestandardization sample and for six age strata. These figuresdemonstrate good internal consistency and are comparableto the values reported for the original Piers-Harris. The age-stratified values are presented to address concerns regardingthe reliability of self-concept scores with younger children.These analyses show that the TOT scale and five of the sixdomain scales maintain good internal consistency through-out the six age ranges. The Popularity (POP) scale demon-strates weaker internal consistency for the youngest children(7- and 8-year-olds; alpha = .60) and also, unexpectedly, forthe oldest adolescents (17- and 18-year-olds; alpha = .62).These findings suggest that the POP scale should be inter-preted cautiously for children in these age ranges.

The standard error of measurement (SEM) statistictranslates the alpha coefficient into practical terms by provid-ing an index of how close an individual test score is likely tobe to the “true” score that would be obtained if there were nomeasurement error. (The formula for calculating SEM for agiven scale is SD √1–r, where SD is the standard deviation andr is the reliability coefficient for that scale.) SEM values forthe Piers-Harris 2 scales are listed in Table 17.

Original Piers-Harris. Table 18 summarizes severalstudies that reported alpha, KR-20, or split-half coefficientsfor the 80-item Total score of the original Piers-Harris.These values, which approach or exceed .90, demonstratethat the original scale has excellent internal consistency forboth younger and older children.

5TECHNICAL PROPERTIES

49

Alpha coefficients for the original six cluster scalesare available from several samples. Piers (1984) reportedalpha coefficients from two samples. The first was a combi-nation of the cluster scale standardization sample (Piers,1984), which consisted of 485 4th through 10th graders, and97 additional children referred from psychiatric clinics. Inthis sample, cluster scale alpha coefficients ranged from .73(Happiness and Satisfaction) to .81 (Behavior), with a meanof .76. The second sample was the 1996 WPS TEST RE-PORT™ sample, a regionally and ethnically diverse pool of1,772 individuals whose Piers-Harris responses were scoredby the WPS TEST REPORT™ service. In this sample, clus-ter scale alpha coefficients ranged from .76 (Happiness andSatisfaction) to .83 (Physical Appearance and Attributes),with a mean of .80. These reliability figures indicate that thecluster scales of the original Piers-Harris have good internalconsistency.

Hattie (1992) reported cluster scale alpha coefficientsfor a sample of 135 Australian students in Grades 10 through12. Alphas ranged from .70 (Happiness and Satisfaction) to.82 (Physical Appearance and Attributes), with a mean of.75. Hattie described similar cluster scale alphas for anothersample of 367 children, with the exception that the internalconsistency of the Happiness and Satisfaction scale wassomewhat low (alpha = .64). This scale also had the lowestalpha coefficient in the three samples described in the previ-ous paragraph. This finding suggests that the Happiness andSatisfaction scale may be more multidimensional than theother Piers-Harris scales, a notion that is supported by sev-eral factor analyses described in the next section.

Test-Retest Reliability of the Original Piers-HarrisTest-retest reliability measures the extent to which

scores for a single individual are consistent over time and

50 Technical Guide

Table 17Piers-Harris 2 Internal Consistency Estimates

Entirestandardization

No. of samplea Age group (Alpha)

Self-Concept scale items Alpha SEM 7–8b 9–10c 11–12d 13–14e 15–16f 17–18g

Total (TOT) 60 .91 3.07 .89 .92 .92 .91 .93 .89

Behavioral Adjustment (BEH) 14 .81 1.27 .75 .84 .81 .81 .81 .76

Intellectual and School Status (INT) 16 .81 1.50 .76 .82 .81 .82 .82 .72

Physical Appearance and Attributes (PHY) 11 .75 1.29 .72 .75 .80 .77 .73 .65

Freedom From Anxiety (FRE) 14 .81 1.46 .77 .82 .82 .82 .84 .80

Popularity (POP) 12 .74 1.37 .60 .72 .80 .79 .78 .62

Happiness and Satisfaction (HAP) 10 .77 1.05 .71 .82 .78 .77 .78 .71aN = 1,387. bn = 188. cn = 231. dn = 277. en = 271. fn = 255. gn = 165.

Table 18Studies Reporting Internal Consistency Coefficients for Piers-Harris Total Score

Study Sample Age or grade Sex n Index r

Center and Ward (1986) Nonclinical Australian Grade 2 Both 183 Split-half .89Grade 3 Both 104 Split-half .91Grades 4, 6, 9 Both 114 Split-half .91

Cooley and Ayres (1988) Mixed nonclinical/special education Grades 6–8 Both 155 Alpha .92Franklin, Duley, et al. (1981) Nonclinical Grades 4–6 Both 180 Alpha .92Lefley (1974) Nonclinical Native American 7–14 years Both 53 Split-half .91Piers (1973) Nonclinical Grade 6 Female 70 KR-20 .88

Male 76 KR-20 .90Grade 10 Female 84 KR-20 .88

Male 67 KR-20 .93Smith and Rogers (1978) Learning disabled 6–12 years Both 206 Alpha .89Winne, Marx, and Taylor (1977) Nonclinical Grades 3–6 Female 42 Alpha .90

Male 61 Alpha .90WPS TEST REPORT™ (1996)a Mixed nonclinical/clinic referred 7–19 years Both 1,772 Alpha .93Yonker, Blixt, and Dinero (1974) Nonclinical Grade 10 Both 208 Alpha .90

aThis sample is a regionally and ethnically diverse pool of individuals from clinical and nonclinical samples whose tests were scored by the WPSTEST REPORT™ service.

across settings. Personality measures such as the Piers-Harris are usually assumed to measure relatively enduringcharacteristics of individuals, and so are expected to producescores that remain stable across time. However, self-conceptmay be less stable among younger children, whose sense ofself is still developing (Harter, 1983). Thus, low test-retestreliability in younger children may be partially due to the in-stability of the underlying construct, rather than measure-ment error per se.

Test-retest reliability data are not available for the Piers-Harris 2 revision. However, a number of studies have investi-gated the test-retest reliability of the original Piers-HarrisTotal score, in both normal children and special populations.Most of the test-retest reliability studies of the original Piers-Harris were completed in the 1960s, 1970s, and early 1980s.The relevant studies are summarized in Table 19. When re-viewing these studies, it is important to note that more hetero-geneous samples are expected to yield higher reliabilitycoefficients, due to greater variance of scores. If for any rea-son a small standard deviation is obtained in a given sample,the test-retest coefficient is expected to be lower. Furthermore,it is not surprising that shorter test-retest intervals are general-ly associated with higher reliability estimates, because there ispresumably less chance that environmental or developmentalchanges affect children’s self-concepts during these shorterintervals. In fact, studies with retest intervals of 6 months orlonger are probably best conceptualized as measuring the sta-

bility of the construct of self-concept over time, rather thantest-retest reliability per se. Table 19 is organized to reflectthis distinction.

Test-retest reliability in general samples. An earlystudy by Piers and Harris (1964) investigated the stability ofthe Piers-Harris in the original standardization sample, usinga 95-item experimental version of the scale and a retest in-terval of 4 months. Reliability coefficients from 3rd, 6th,and 10th graders were .72, .71, and .72, respectively. Thesereliability estimates were deemed satisfactory by the au-thors, especially given the relatively long retest interval andthe fact that the scale was still in the development stage. The80-item Piers-Harris, though shorter than the experimentalversion, was shown to have marginally better stability forboth 2-month (r = .77) and 4-month (r = .77) retest intervals(Wing, 1966). Additional studies of nonclinical students re-port reliability coefficients ranging from .65 to .81 over 2- to5-month retest intervals (McLaughlin, 1970; Platten &Williams, 1979, 1981; Shavelson & Bolus, 1982).

Hattie (1992) reported a test-retest study for the Piers-Harris Total score and the six cluster scales. The study useda sample of 135 Australian students in Grades 10 through12. The test-retest interval was 4 weeks. The reliability co-efficients were as follows: Total, .87; Behavior, .80; In tel -lect ual and School Status, .84; Physical Appearance andAttributes, .88; Anxiety, .80; Popularity, .80; Happiness andSatisfaction, .65.

Chapter 5 Technical Properties 51

Table 19Studies Reporting on Test-Retest Reliability and Construct Stability for Piers-Harris Total Score

RetestStudy Sample Age or grade Sex n interval r

Test-retest reliabilitya

Alban Metcalfe (1981) British 11–20 years Both 182 2 weeks .69Lefley (1974) American Indian 7–14 years Both 40 10 weeks .73McLaughlin (1970) Private school Grade 5 Male 67 5 months .75

Grade 6 Male 98 5 months .72Grade 7 Male 69 5 months .71

Piers and Harris (1964)b Public school Grade 3 Both 56 4 months .72Grade 6 Both 66 4 months .71Grade 10 Both 60 4 months .72

Platten and Williams (1979) Mixed ethnic groups Grades 4–6 Both 159 10 weeks .65Platten and Williams (1981) Both 173 10 weeks .75Querry (1970) Normal speech Grades 3–4 Both 10 3–4 weeks .86

Mild articulation disorders Both 10 3–4 weeks .96Moderate articulation disorders Both 10 3–4 weeks .83

Shavelson and Bolus (1982) Public school Grades 7–8 Both 99 5 months .81Tavormina (1975) Chronic medical illness M = 12 years Both 94 3–4 weeks .80Wing (1966) Public school Grade 5 Both 244 2 months .77

4 months .77

Construct stabilityc

Henggeler and Tavormina (1979) Mexican American M = 10.5 years Both 12 1 year .51Smith and Rogers (1977) Learning disability 6–12 years Both 89 6 months .62Wolf (1981) Mental retardation/emotional disturbance 11–16 years Both 39 8 months .42

aRetest intervals of less than 6 months. bBased on 95-item experimental version of the Piers-Harris. cRetest intervals of 6 months or more.

Test-retest reliability in special populations. Manystudies have reported test-retest reliability estimates forsamples of minority and special-needs students. These stud-ies have differed in terms of the sample characteristics (e.g.,age, sex, and group membership) and retest intervals. Somestudies have even modified the administration procedures orthe instrument itself. Thus, it is not unexpected that the reli-ability coefficients for these studies vary considerably.

Minority students. The stability of the Piers-Harris inpopulations of diverse ethnic or national backgrounds can beascertained from three studies. Lefley (1974) found a reliabil-ity coefficient of .73 for a sample of Native American studentstested over a 10-week interval. Henggeler and Tavormina(1979) obtained 1-year retest data from 12 Mexican Americanmigrant children. The reported reliability coefficient of .51 isbest understood as representing construct stability, as opposedto test-retest reliability. Viewed in this light, the coefficient isactually rather high, given the long retest interval and smallsample size. Finally, Alban Metcalfe (1981) reported a2-week test-retest coefficient of .69 in a sample of studentsfrom Northern England. This study used a modified Piers-Harris, with “Americanisms” removed and the response for-mat altered to a 5-point Likert-type scale.

Students with special needs. Studies of students withspecial needs have included investigations of children withmental retardation, children with learning and speech dis-abilities, and children with chronic medical illness. Querry(1970) compared the self-concepts of 3rd and 4th graderswith normal speech, mild articulation disorders, and moder-ate articulation disorders. The reported test-retest reliabilitycoefficients were .86, .96, and .83, respectively, which isnotable considering the small sample size in each group(n = 10). Tavormina (1975) obtained a stability coefficientof .80 with a sample of children with chronic illnesses. Theselatter two studies were based on relatively short (3 to 4 week)re test intervals. In a study of students with learning disabili-ties (ages 6 to 12), Smith and Rogers (1977) reported a reli-ability coefficient of .62 over a retest interval of 6 months. Ina study of 39 students identified as having mental retardationor emotional disturbances, Wolf (1981) reported a stabilitycoefficient of .42 over an 8-month interval. Again, this studyis probably best construed as an example of construct stabil-ity rather than test-retest reliability. The long retest interval,small sample size, and relatively low variance of scores(compared with the Piers-Harris normative sample) proba-bly all contributed to the low coefficient in this study.

Unreliability due to random responding. Wylie(1974) has suggested that low scores on self-concept testsmay be less reliable than higher scores, particularly when thelow scores are generated by younger children. With youngerchildren, it is possible that inability to read or understand theitems could lead to random responding. A randomly an-swered questionnaire tends to produce a score at or near thescale’s midpoint. This is likely to result in a low self-conceptscore, in normative terms, simply because most self-conceptscales produce negatively skewed distributions, with scoresaccumulating disproportionately above the scale’s midpoint.

Random response patterns need to be considered wheninterpreting the results of test-retest reliability studies.Random responding tends to result in inconsistent item re-sponses between two testing occasions, which is a hallmarkof poor reliability. However, children who respond randomlyare likely to have low self-concept scores on both occasions,for the reasons described previously. This creates a situationthat in fact is emblematic of low test-retest reliability, butnevertheless generates a spuriously high test-retest correla-tion (because both scores are in the low range). This situationmight be interpreted as reflecting a substantial relationshipbetween the two testing occasions when actually the onlycommon element was the randomness of responding.

These speculations raise two empirical questions: (a)Do Piers-Harris scores below the mean exhibit more item in-stability across testing occasions than scores above the mean?and (b) if this differential instability exists, is it due to randomresponding? The Smith and Rogers (1977) study providesdata relevant to these questions. Smith and Rogers adminis-tered the Piers-Harris to 89 children (aged 6 to 12) with sig-nificant learning deficiencies. Extra care was taken to ensurethat the children understood the items. The sample was splitinto three groups based on initial Piers-Harris Total score(high, middle, or low). The groups did not differ significantlyin age or IQ. The sample was given the Piers-Harris again 6months later. An index of instability was determined for eachchild by calculating the number of item inconsistencies fromthe first testing to the second testing. Results indicated thatthe index of instability for the high self-concept group wassignificantly less than for the middle or low groups, but thatthe latter two groups did not differ significantly.

Because those with low and middle scores did not dif-fer in item-response stability, Smith and Rogers (1977) con-cluded that random responding alone could not account forlow test-retest reliability among low scorers. If random re-sponding were the cause of low reliability, then the middle-scoring group, which would have included fewer randomresponse sets, would have showed less item instability thanthe low-scoring group. As an alternative explanation fortheir findings, Smith and Rogers proposed that for youngerchildren, the point in time at which self-concept stabilizes isa function of the favorableness of self-concept. Under thisview, younger children with high self-regard demonstratestability of self-concept earlier than do same-age childrenwith low self-regard. Once self-concept has become rela-tively stable (as in older children and adults), no differentialitem instability for those with high, middle, and low self-concept scores would be expected. According to this theory,then, greater variability in low-scoring children should beinterpreted as reflecting uncertain and poorly defined self-image rather than inadequate test-retest reliability. Clearly,more research is needed to test these hypotheses.

Validity

Validity refers to a test’s ability to measure accuratelythose psychological characteristics that it purports to measure.

52 Technical Guide

Validity is a multidimensional concept that can be dividedinto several types, each of which plays a different role in es-tablishing the usefulness and accuracy of a test (Anastasi,1988). Content validity addresses the question of whetherthe test’s item content adequately samples the behavior thatis being measured. A second type of validity, construct va-lidity, refers to how well the test performs in measuring atheoretical psychological characteristic (e.g., introversion,neuroticism). Finally, criterion validity involves how wellthe test performs in predicting an individual’s performanceor status in other activities (e.g., school achievement, re-sponse to psychiatric treatment).

Since the introduction of the original Piers-Harris inthe early 1960s, researchers have produced a large body ofevidence supporting the measure’s validity. The validationprocess for the Piers-Harris 2 is based on this existing liter-ature. In addition, new data concerning construct validityhave been collected as part of the standardization study forthe Piers-Harris 2.

This section begins by briefly examining the contentvalidity of the Piers-Harris 2, in terms of the evolution andrefinement of the original Piers-Harris item set. Contentvalidity is the least important aspect of the validity of thePiers-Harris 2, simply because self-concept is by definitiona theoretical entity, and thus the validity of the Piers-Harris 2is more appropriately determined by construct validationmethods. In addition, the Pier-Harris 2 is frequently used asan outcome measure, which highlights the importance of es-tablishing its criterion validity. The majority of this sectionis therefore devoted to construct and criterion validity, de-scribing first the recent studies contributing to the Piers-Harris 2 revision, followed by a review of validity researchpertaining to the original measure.

Content Validity of the Original andRevised Instruments

Piers-Harris 2. The Piers-Harris 2 contains 60 items,20 fewer than the item set of the original measure. The ques-tion arose as to whether this item reduction had any impacton the content validity of the Piers-Harris 2. This issue con-cerns only the TOT score, as the Piers-Harris 2 domainscales are almost identical to their counterparts in the origi-nal measure. To determine if the item reduction would af-fect content validity, a clinical judge compared the deleteditems (see Appendix E) with the retained items (see Table4). Sixteen of the deleted items were judged to have ade-quate content overlap with the retained items, so it was con-cluded that deleting these items would not result in anoverall loss of content coverage in the Piers-Harris 2. Thejudge identified four deleted items that had relatively littlecontent overlap with the retained items. These four itemswere: “I am good at making things with my hands,” “I candraw well,” “I am good in music,” and “I sleep well atnight.” These items refer to specific abilities and attributesrather than more general facets of self-concept (e.g., “I am agood person”). Therefore, it was decided that the effect ofdeleting them on the overall content sampling of the Piers-

Harris 2 would be relatively small, and that the content valid-ity of the measure would not be threatened by going aheadwith the planned item reduction.

Original Piers-Harris. The items of the originalPiers-Harris were written with the goal of maximizing con-tent validity. The universe of content to be sampled was de-fined as qualities that children reported liking or dislikingabout themselves (Jersild, 1952). Children’s statementswere grouped into the following categories: (a) physicalcharacteristics and appearance; (b) clothing and grooming;(c) health and physical well-being; (d) home and family; (e) enjoyment of recreation; (f) ability in sports and play; (g) academic performance and attitudes toward school; (h) intellectual abilities; (i) special talents (music, arts); (j) “Just Me, Myself”; and (k) personality characteristics,inner resources, and emotional tendencies. As detailed inchapter 4, an initial pool of 164 pilot items was developed toreflect all of these categories. This pool was reduced to 80items by eliminating items with poor ability to discriminatebetween high and low scores on the entire item set. The reduceditem set contained items from all of the original categories.

The original factor analysis (Piers, 1963) identified sixfactors that became the cluster scales: Behavior, Intellectualand School Status, Physical Appearance and Attributes,Anxiety, Popularity, and Happiness and Satisfaction. Thesefactors collapsed several of Jersild’s (1952) categories andemphasized items from the two most general categories(“Just Me, Myself” and personality characteristics, inner re-sources, and emotional tendencies). These general cate-gories presumably are a better reflection of a child’s overallself-concept than narrower categories (e.g., special talentsor enjoyment of recreation).

Construct Validity of the Piers-Harris 2The Piers-Harris 2 standardization study provided two

kinds of evidence related to the construct validity of the re-vised instrument. The study allowed detailed examination ofthe instrument’s structural characteristics, which refers tothe intercorrelations and item composition of the Piers-Harris2 domain scales. Determining the interrelatedness of the do-main scales helps establish whether they can be viewed asmeasuring separate components of overall self-concept. Inaddition, concurrent data was collected on other psycho-logical tests from subsamples of the Piers-Harris 2 standard-ization sample. This data enables assessment of convergentvalidity, or the extent to which the Piers-Harris 2 correlateswith measures of similar psychological constructs.

Structural characteristics. Interscale correlationanalysis and factor analysis are the two methods used to il-luminate the structural characteristics of the Piers-Harris 2.

Interscale correlations. Table 20 presents interscalecorrelation coefficients for the Piers-Harris 2 standardizationsample. Most of the scales exhibit correlations with eachother in the moderate to high moderate range. Inter scale cor-relations of this magnitude are to be expected for several rea-sons. First of all, each scale shares items with at least twoother scales. The magnitude of the interscale correlation

Chapter 5 Technical Properties 53

coefficient reflects the number of shared items betweenscales. For example, the FRE and HAP scales, which sharefour items, correlate at r = .66. In contrast, BEH and POP,which do not share any items, have a much weaker associa-tion (r = .30). Furthermore, the theory underlying the Piers-Harris specifies that a child’s general sense of self-worthshould influence his or her self-appraisals in specific areasof functioning. This suggests that the factor-analytically de-rived subscales of the test, which are intended to measuredistinct dimensions of self-concept, should neverthelessshare variance with one another.

Additional findings support the notion that the domainscales represent separate but interrelated aspects of self-concept. First, in all cases, the interscale correlations arelower than the scale reliabilities (see Table 17). This indi-cates that individual items are related more strongly to otheritems on the same domain scale than to items on other scales(with the exception of overlapping items). Second, as Table20 shows, each domain scale correlates more strongly withthe TOT score than with any of the other content scales. Thisdemonstrates that each domain scale is a better index of gen-eral self-concept than of the particular components of self-concept measured by the other domain scales.

Factor analysis. An exploratory common factor anal-ysis with oblimin rotation was conducted using the Piers-Harris 2 standardization sample data. The common factorapproach was selected because it allows for sources of vari-ance (e.g., measurement error) other than the extracted fac-tors. Oblimin rotation was chosen because it assumescorrelated factors, which is a theoretically and empiricallyreasonable assumption for this measure (see the discussionin the previous section). Table 21 presents the factor load-ings (item-factor correlations), with items organized by do-main scale.

The factor analysis yielded six factors with eigenval-ues greater than 1. The first factor was weighted with itemsrepresenting feelings of happiness and perceptions that oneis important and valued by others. The second factor reflect-ed endorsement of troublesome behaviors at school and athome. The third factor appeared to represent freedom fromanxiety, worry, and nervousness. The fourth factor reflectedperceptions of being good at schoolwork and fitting in well

at school. The fifth factor represented dissatisfaction withone’s physical appearance and personal attributes. The sixthfactor reflected a perception that one has many friends,makes new friends easily, and is well liked by others. Itshould be noted that the first, third, fourth, and sixth factorsrepresented qualities that are positively correlated with self-esteem, and the second and fifth factors represented qualitiesthat are negatively correlated with self-esteem. Thus, a childwith a high self-concept score would be expected to haverelatively high scores on Factors I, III, IV, and VI, and rela-tively low scores on Factors II and V.

There appears to be reasonably good correspondencebetween the results of the factor analysis and the item assign-ments for the Piers-Harris 2 domain scales. The BEH, INT,FRE, and POP scales map relatively cleanly onto Factors II,IV, III, and VI, respectively. The situation is a bit more com-plex for the other two domain scales. On the PHY scale, theitems that reflect concerns with physical appearance mapcleanly onto Factor V. However, PHY items that representother personal attributes (e.g., Item 9, “I am a leader ingames and sports,” or Item 15, “I am strong”) do not load onFactor V or any other factor. The HAP scale appears to bebifactorial. There is a strong loading on Factor I of itemsrepresenting feelings of happiness, importance, and beingvalued by others. Other HAP items having to do with satis-faction with one’s appearance load on Factor V, reflectingsome of the item overlap between HAP and PHY.

It is worth noting that in this factor analysis, itemswere recoded to conform to the Piers-Harris 2 scoring sys-tem (i.e., “1” is coded for the response in the direction ofpositive self-concept, “0” for the alternative response). Thisrecoding tends to obliterate the original distinction betweenpositive and negative item phrasing. For example, a posi-tively phrased item (e.g., Item 60, “I am a good person”) iscoded “1” for a yes response, and a negatively phrased item(e.g., Item 36, “I hate school”) is coded “1” for a no re-sponse. Closer inspection of the factor analysis, however,does reveal some effects of item phrasing on the factor struc-ture of the Piers-Harris 2. In particular, Factors II and III arecomposed primarily of negatively phrased items; Factors IVand V contain mostly positively phrased items; and Factors Iand VI are evenly mixed.

54 Technical Guide

Table 20Interscale Correlations in the Piers-Harris 2 Standardization Sample

TOT BEH INT PHY FRE POP HAP

Total (TOT) -

Behavioral Adjustment (BEH) .73 -

Intellectual and School Status (INT) .84 .64 -

Physical Appearance and Attributes (PHY) .76 .34 .65 -

Freedom From Anxiety (FRE) .79 .42 .52 .50 -

Popularity (POP) .75 .30 .50 .66 .64 -

Happiness and Satisfaction (HAP) .81 .53 .60 .69 .66 .55 -

Note. N = 1,387.

55

Table 21Factor Loadings for Piers-Harris 2 Item Responses in Standardization Sample

I II III IV V VI

Behavioral Adjustment (BEH)

12. I am well behaved in school. (Y) –.44

13. It is usually my fault when something goes wrong. (N) –.41

14. I cause trouble to my family. (N) –.49

18. I am good in my schoolwork. (Y) .56

19. I do many bad things. (N) –.62

20. I behave badly at home. (N) –.56

27. I often get into trouble. (N) –.62

30. My parents expect too much of me. (N)

36. I hate school. (N) .45

38. I am often mean to other people. (N) –.47

45. I get into a lot of fights. (N) –.50

48. My family is disappointed in me. (N) –.47

58. I think bad thoughts. (N) –.46

60. I am a good person. (Y) .53

Intellectual and School Status (INT)

5. I am smart. (Y) .51

7. I get nervous when the teacher calls on me. (N) .51

12. I am well behaved in school. (Y) –.44

16. I am an important member of my family. (Y) .49

18. I am good in my schoolwork. (Y) .56

21. I am slow in finishing my schoolwork. (N)

22. I am an important member of my class. (Y) .44

24. I can give a good report in front of the class. (Y) .51

25. In school I am a dreamer. (N)

26. My friends like my ideas. (Y) .49

34. I often volunteer in school. (Y) .45

39. My classmates in school think I have good ideas. (Y) .49 .50

43. I am dumb about most things. (N) .41

50. When I grow up, I will be an important person. (Y) .42

52. I forget what I learn. (N) .50

55. I am a good reader. (Y) .46

Physical Appearance and Attributes (PHY)

5. I am smart. (Y) .51

8. My looks bother me. (N) –.58

9. I am a leader in games and sports. (Y)

15. I am strong. (Y)

26. My friends like my ideas. (Y) .49

33. I have nice hair. (Y) –.55

39. My classmates in school think I have good ideas. (Y) .49 .50

44. I am good-looking. (Y) –.65

46. I am popular with boys. (Y)

49. I have a pleasant face. (Y) –.63

54. I am popular with girls. (Y) .42

continued on next page . . .

56

Table 21 (continued)Factor Loadings for Piers-Harris 2 Item Responses in Standardization Sample

I II III IV V VI

Freedom From Anxiety (FRE)

4. I am often sad. (N) .44

6. I am shy. (N)

7. I get nervous when the teacher calls on me. (N) .51

8. My looks bother me. (N) –.58

10. I get worried when we have tests in school. (N) .50

17. I give up easily. (N)

23. I am nervous. (N) .55

29. I worry a lot. (N) .56

31. I like being the way I am. (Y) –.54

32. I feel left out of things. (N) .48 –.46

35. I wish I were different. (N) .42 –.58

40. I am unhappy. (N) .54

56. I am often afraid. (N) .54

59. I cry easily. (N) .44

Popularity (POP)

1. My classmates make fun of me. (N) .41

3. It is hard for me to make friends. (N) .56

6. I am shy. (N)

11. I am unpopular. (N) .52

32. I feel left out of things. (N) .48 –.46

37. I am among the last to be chosen for games and sports. (N) .51

39. My classmates in school think I have good ideas. (Y) .49 .50

41. I have many friends. (Y) .43 .63

47. People pick on me. (N) .52

51. In games and sports, I watch instead of play. (N)

54. I am popular with girls. (Y) .42

57. I am different from other people. (N)

Happiness and Satisfaction (HAP)2. I am a happy person. (Y) .63

8. My looks bother me. (N) –.58

28. I am lucky. (Y)

31. I like being the way I am. (Y) –.54

35. I wish I were different. (N) .42 –.58

40. I am unhappy. (N) .54

42. I am cheerful. (Y) .60

49. I have a pleasant face. (Y) –.63

53. I am easy to get along with. (Y)

60. I am a good person. (Y) .53

Note. N = 1,387. Principal axis extraction method with oblimin rotation. Loadings are item-factor correlations. Loadings of less than .40 are not displayed. Letter in parentheses indicates response scored as positive self-concept (Y = yes, N = no).

Overall, the factor analysis supports the use of thePiers-Harris 2 domain scales to measure distinct aspects ofself-concept. The factor solution includes four factors whoseitem composition corresponds closely to the item assign-ments for the BEH, INT, FRE, and POP domain scales. Theremaining two factors share items from the HAP and PHYdomain scales. The results of the current analysis are consis-tent with many of the prior factor analyses performed on theoriginal Piers-Harris, which are described in the next section.

The structure of the domain scales is further clarifiedby item-total correlations for the self-concept scales, whichare presented in Table 22. These correlations are correctedfor item-scale assignments, meaning that an item’s score issubtracted from its assigned scale score before the corre-lation between that item and that scale is calculated. Cor re -lations between items and their assigned domain scales areexpected to be relatively homogenous and higher than corre-lations between those items and other domain scales. Table22 confirms these expectations. For items and their assigneddomain scale, correlations range from .25 to .59 for BEH,.25 to .53 for INT, .28 to .58 for PHY, .27 to .51 for FRE, .17to .54 for POP, and .32 to .52 for HAP. Correlations betweenitems and other domain scales to which they do not belongare lower overall, ranging from –.01 to .50.

Thus, although significant correlations with other do-main scales do exist, 52 of the 60 items have their highestcorrelation with their assigned domain scale. Correlationsbetween items and the TOT score range from .17 to .56, sug-gesting that all items are significantly associated with thisgeneral measure of self-concept.

Convergent validity. Convergent validity of Piers-Harris 2 was measured by correlating the test with concur-rent measures of anger and aggressive attitudes, as well aswith measures of psychological symptoms.

Correlations with measures of anger and aggressiveattitudes. Table 23 presents correlations between the Piers-Harris 2 and two measures of anger and aggressive attitudes:the Attitudes Toward Guns and Violence Questionnaire(AGVQ; Shapiro, 2000) and the Aggression Questionnaire(AQ; Buss & Warren, 2000). The AGVQ is a measure of at-titudes concerning guns, physical aggression, and inter-personal conflict. Because the AGVQ scores correlatestrongly with interpersonal behavior problems, aggressivebehavior, and aggression-related emotions, the scores wouldbe expected to have an inverse relationship with self-esteem.As Table 23 demonstrates, the Piers-Harris 2 TOT score andfour of the six domain scale scores show significant negativecorrelations with AGVQ Total score. The predicted relation-ship holds most strongly for the BEH scale (r = –.46), whichcontains items that tap aggressive attitudes (e.g., Item 38, “Iam often mean to other people,” and Item 45, “I get into a lotof fights”). These Piers-Harris 2 scales show the expectednegative associations with one of the AGVQ subscales(Aggressive Response to Shame), which measures compo-nents of anger that are related to self-concept, including sen-sitivity to disrespect from others, and the belief that violencecan repair damage to self-esteem (Shapiro, 2000). Corre -

lation coefficients between the Piers-Harris 2 scales and theAggressive Response to Shame subscale ranged from –.14(HAP) to –.45 (BEH).

The AQ is a measure of aggressive attitudes and angrybeliefs. Table 23 shows that the predicted negative relation-ships between the Piers-Harris 2 scales and the AQ held for thePiers-Harris 2 TOT score and three of the six domain scales.As with the AGVQ, the Piers-Harris 2 BEH scale demon-strated the strongest correlation with AQ Total score (r = –.32).

Correlations with measures of psychological symp-toms. The My Worst Experience Scale (MWES; Hyman &Snook, 2002) is a measure of Posttraumatic Stress Disordersymptoms in children. The MWES taps many symptoms ofgeneral emotional distress that would be expected to corre-late inversely with a measure of self-concept. Table 24shows correlations between Piers-Harris 2 scales andMWES scales. As with many large-sample correlation ma-trices, the majority of the correlation coefficients are statis-tically significant. Of greater interest than statisticalsignificance is the magnitude of the correlation coefficients,which indexes the strength of association between thevarious scales. In Table 24, each of the larger coefficients(r ≥ .25 in absolute value) is displayed in bold type.

The table shows the predicted negative relationshipsbetween the MWES Total score and the three Piers-Harris 2domain scales that are the strongest indicators of emotionaland behavioral problems (BEH, FRE, HAP). In addition,each of these three domain scales has a relatively strong neg-ative association (r ≥ .30 in absolute value) with three MWESscales that are close matches in terms of scale content(Oppositional Conduct, Hypervigilance, and Hopelessness,respectively). Finally, there are relatively few MWES itemsthat assess the quality of a child’s interpersonal functioning.Thus, the MWES scales would be predicted to have weakercorrelations with POP than with other Piers-Harris 2 scales,and Table 24 shows that this is indeed the case.

The Overeating Questionnaire (OQ; O’Donnell &Warren, in press) is a self-report instrument that measuresthoughts and attitudes related to obesity. Its relevance to thevalidity of the Piers-Harris 2 comes from the fact that obesi-ty, like other eating disorders, is often associated with fun-damental deficits in self-esteem. Table 25 presentscorrelations between Piers-Harris 2 scales and OQ scales,with larger coefficients (r ≥ .25 in absolute value) displayedin bold type. The table shows that all of the Piers-Harris 2scales correlate relatively strongly in the expected negativedirection with the OQ Affective Disturbance scale, which isa general measure of emotional problems. In addition, spe-cific scale-to-scale relationships are relatively strong and inthe predicted direction. For example, the Piers-Harris 2 PHYscale is related to the OQ Body Image scale, and the Piers-Harris 2 POP scale shows an inverse correlation with the OQSocial Isolation scale.

Construct Validity of the Original Piers-HarrisFactor structure. The conceptual underpinnings of

the original Piers-Harris have been investigated by a series

Chapter 5 Technical Properties 57

58

Table 22Piers-Harris 2 Corrected Item-Total Correlations

Intellectual Physical Happinessand Appearance Freedom and

Behavioral School and From Satis-Total Adjustment Status Attributes Anxiety Popularity faction

1. My classmates make fun of me. .37 .21 .25 .30 .32 .38 .322. I am a happy person. .43 .33 .33 .34 .28 .29 .473. It is hard for me to make friends. .36 .15 .22 .31 .33 .47 .304. I am often sad. .48 .36 .33 .29 .47 .37 .455. I am smart. .47 .37 .53 .35 .27 .26 .376. I am shy. .19 –.01 .11 .17 .27 .22 .127. I get nervous when the teacher calls on me. .35 .16 .25 .22 .44 .31 .238. My looks bother me. .41 .20 .26 .32 .44 .34 .389. I am a leader in games and sports. .26 .03 .23 .29 .21 .30 .21

10. I get worried when we have tests in school. .26 .15 .22 .12 .38 .21 .1211. I am unpopular. .33 .10 .23 .36 .28 .45 .2612. I am well behaved in school. .31 .44 .34 .19 .09 .08 .2613. It is usually my fault when something

goes wrong. .39 .37 .31 .22 .33 .25 .3314. I cause trouble to my family. .40 .46 .29 .20 .29 .24 .3215. I am strong. .33 .17 .27 .32 .26 .25 .2916. I am an important member of my family. .42 .35 .38 .35 .23 .25 .4317. I give up easily. .40 .31 .34 .27 .39 .24 .3318. I am good in my schoolwork. .46 .43 .52 .32 .27 .21 .3419. I do many bad things. .38 .59 .39 .16 .19 .10 .2920. I behave badly at home. .38 .51 .33 .17 .25 .17 .2821. I am slow in finishing my schoolwork. .41 .30 .39 .28 .33 .30 .2622. I am an important member of my class. .42 .25 .44 .45 .26 .34 .3523. I am nervous. .39 .19 .28 .23 .49 .37 .2824. I can give a good report in front of the class. .40 .27 .46 .34 .26 .24 .2925. In school I am a dreamer. .31 .31 .31 .11 .24 .17 .1926. My friends like my ideas. .46 .28 .44 .48 .27 .42 .3827. I often get into trouble. .37 .56 .36 .15 .20 .13 .2628. I am lucky. .31 .17 .28 .31 .22 .21 .3229. I worry a lot. .41 .22 .27 .25 .51 .34 .3230. My parents expect too much of me. .24 .25 .18 .08 .24 .16 .1931. I like being the way I am. .48 .30 .35 .37 .45 .35 .5132. I feel left out of things. .51 .27 .36 .36 .51 .46 .4333. I have nice hair. .40 .25 .35 .45 .23 .28 .4434. I often volunteer in school. .28 .27 .36 .25 .11 .12 .2435. I wish I were different. .48 .32 .35 .37 .50 .34 .4636. I hate school. .31 .37 .40 .15 .14 .10 .2037. I am among the last to be chosen

for games and sports. .43 .21 .29 .38 .37 .49 .3438. I am often mean to other people. .33 .43 .27 .16 .19 .16 .2539. My classmates in school think

I have good ideas. .49 .29 .50 .50 .27 .36 .3940. I am unhappy. .56 .43 .40 .35 .45 .42 .5241. I have many friends. .48 .19 .35 .50 .35 .54 .4542. I am cheerful. .45 .34 .41 .40 .26 .31 .4943. I am dumb about most things. .53 .46 .50 .36 .37 .35 .4044. I am good-looking. .44 .20 .36 .58 .30 .37 .5045. I get into a lot of fights. .25 .42 .22 .07 .11 .11 .17

continued on next page . . .

of factor-analytic studies. These studies have raised twoissues regarding the factor composition of the scale. First,factor-analytic studies have examined the structure of theunderlying dimensions of self-concept. Second, assumingmultiple dimensions do exist, studies have attemptedto replicate these dimensions across various age and eth-nic groups.

As noted in chapter 4, Piers (1963) conducted the ini-tial factor analysis of the scale with a sample of 457 sixthgraders. She used the principal components method withvarimax rotation to identify six interpretable factors that to-gether accounted for 42% of the variance in item responses.These factors, which became the original Piers-Harris clus-

ter scales, were labeled Behavior, Intellectual and SchoolStatus, Physical Appearance and Attributes, Anxiety, Pop -ularity, and Happiness and Satisfaction.

Piers (1973) conducted a second factor analysis with146 sixth graders and 151 tenth graders, using the samemethod as the initial study. The sixth-grade analysis con-firmed the original six factors and identified three addition-al factors: Self-Worth, Adjustment, and a Strength versusTimidity factor related to the Anxiety factor. In the tenth-grade sample, all original factors were confirmed with theexception of the Physical Appearance and Attributes factor,which appeared to be collapsed into the Intellectual andSchool Status factor. The study identified two additionalfactors: Outgoing Activity and Leadership, and a secondBe havior factor.

W. B. Michael, Smith, and J. J. Michael (1975) at-tempted to replicate Piers’s factor structure using three sam-ples: 299 elementary school pupils, 302 junior high schoolstudents, and 300 high school students. The study utilized a principal components method with varimax rotation. Allthree samples confirmed the Behavior factor and PhysicalAppearance and Attributes factor. Two samples (elemen-tary school and high school) confirmed the Intellectual and School Status factor and the Anxiety factor. A Self-Dis satisfaction factor involving Anxiety was found in thejunior high school sample. The Popularity factor and theHap pi ness and Satisfaction factor were not confirmed in theelementary school or high school samples. In the junior highschool sample, support was found for the Popularity factorand for a factor labeled Self-Contentment, which resembledthe Happiness and Satisfaction factor. Taken together, these

Chapter 5 Technical Properties 59

Table 22 (continued)Piers-Harris 2 Corrected Item-Total Correlations

Intellectual Physical Happinessand Appearance Freedom and

Behavioral School and From Satis-Total Adjustment Status Attributes Anxiety Popularity faction

46. I am popular with boys. .22 .02 .18 .28 .15 .32 .2147. People pick on me. .39 .21 .23 .31 .34 .48 .3048. My family is disappointed in me. .44 .49 .33 .24 .34 .23 .4149. I have a pleasant face. .50 .30 .40 .57 .33 .39 .4950. When I grow up, I will be an

important person. .33 .22 .33 .34 .17 .19 .3351. In games and sports, I watch instead of play. .31 .16 .20 .24 .31 .28 .2452. I forget what I learn. .43 .43 .48 .24 .33 .19 .2753. I am easy to get along with. .37 .32 .31 .30 .20 .28 .3454. I am popular with girls. .22 .01 .17 .30 .14 .32 .1955. I am a good reader. .34 .33 .41 .26 .17 .16 .2456. I am often afraid. .38 .22 .23 .21 .49 .33 .2857. I am different from other people. .17 .10 .11 .10 .18 .17 .1158. I think bad thoughts. .41 .46 .38 .20 .30 .17 .3459. I cry easily. .26 .13 .12 .15 .40 .23 .2260. I am a good person. .44 .39 .39 .34 .24 .25 .44

Note. Coefficients in bold type indicate that item’s domain scale assignments. The sign is omitted from all correlation coefficients.

Table 23Correlations Between Piers-Harris 2 Scores andMeasures of Anger and Aggressive Attitudes

AGVQa AQb

Piers-Harris 2 scale Total Total

Total (TOT) –.29** –.16*

Behavioral Adjustment (BEH) –.46** –.32**

Intellectual and School Status (INT) –.31** –.13*

Physical Appearance and Attributes (PHY) –.08 .04

Freedom From Anxiety (FRE) –.16** –.14*

Popularity (POP) –.08 .01

Happiness and Satisfaction (HAP) –.21** –.08aAttitude Toward Guns and Violence Questionnaire. N = 414, ages 9 to18. bAggression Questionnaire. N = 294, ages 9 to 18.

*p < .05. **p < .005.

60 Technical Guide

Table 24Correlations Between Piers-Harris 2 and My Worst Experience Scale (MWES) Scores

Piers-Harris 2 scale

Intellectual Physical Happinessand Appearance Freedom and

Behavioral School and From Satis-MWES scale Total Adjustment Status Attributes Anxiety Popularity faction

MWES Total –.30 –.25 –.23 –.16 –.29 –.16 –.27

Impact of the Event –.28 –.23 –.21 –.15 –.32 –.16 –.25

Re-experience of the Trauma –.18 –.16 –.12 –.08 –.21 –.12 –.18

Avoidance and Numbing –.24 –.17 –.16 –.13 –.27 –.16 –.25

Increased Arousal –.26 –.25 –.19 –.13 –.25 –.13 –.23

Depression –.22 –.12 –.13 –.12 –.27 –.18 –.24

Hopelessness –.33 –.25 –.27 –.26 –.30 –.19 –.32

Oppositional Conduct –.29 –.38 –.29 –.18 –.16 –.08 –.21

Hypervigilance –.26 –.24 –.18 –.11 –.30 –.10 –.27

Dissociation and Dreams –.32 –.26 –.27 –.19 –.29 –.20 –.25

General Maladjustment –.25 –.30 –.27 –.10 –.20 –.11 –.13

Somatic Symptoms –.17 –.16 –.10 –.02 –.25 –.07 –.19

Note. N = 150. Values for which |r | ≥ .25 are in bold type.

Values for which |r | < .16 are nonsignificant. For .16 ≤ |r | < .24, p < .05. For |r | ≥ .24, p < .005.

Table 25Correlations Between Piers-Harris 2 Scores and Overeating Questionnaire (OQ) Scores

Piers-Harris 2 scale

Intellectual Physical Happinessand Appearance Freedom and

Behavioral School and From Satis-OQ scale Total Adjustment Status Attributes Anxiety Popularity faction

Overeating –.27 –.21 –.25 –.12 –.22 –.26 –.18

Undereating –.23 –.10 –.09 –.19 –.35 –.22 –.25

Craving –.16 –.15 –.13 –.07 –.16 –.14 –.06

Expectations –.21 –.07 –.19 –.19 –.18 –.21 –.14

Rationalizations –.29 –.16 –.23 –.24 –.25 –.26 –.27

Motivation –.16 0 –.05 –.14 –.29 –.21 –.15

Health Practices .26 .24 .28 .21 .10 .18 .12

Body Image .34 .09 .14 .43 .36 .38 .42

Social Isolation –.51 –.14 –.36 –.46 –.51 –.58 –.46

Affective Disturbance –.50 –.25 –.28 –.38 –.56 –.46 –.43

Note. N = 251. Values for which |r | ≥ .25 are in bold type.

Values for which |r | < .13 are nonsignificant. For .13 ≤ |r | < .18, p < .05. For |r | ≥ .18, p < .005.

results strongly support the first three factors identified byPiers (Behavior, Intellectual and School Status, and PhysicalAppearance and Attributes). The Anxiety factor was alsoconfirmed in the three samples, although more diffusely, butthe Popularity factor and Happiness and Satisfaction factorwere confirmed in only one sample.

The factor structure of the Piers-Harris was studied ina mixed-ethnicity (Anglo-American and African American)sample by Wolf, Sklov, Hunter, Webber, and Berenson

(1982). The sample consisted of 348 students, aged 10 to 17,from a rural southern community. A principal componentsanalysis identified seven interpretable factors accounting for38.7% of the variance. These factors included the six identi-fied by Piers and an additional factor, labeled Aggression.Some ethnic and gender differences were identified in thefactor structure. Chief among these were that Caucasiansscored higher than African Americans on the Intellectualand School Status factor, and females scored lower than

males on the Behavior factor, the Intellectual and SchoolStatus factor, and the Anxiety factor.

Another factor analysis used a sample of 170 studentswith educable mental retardation (EMR; Rich, Barcikowski,& Witmer, 1979). The principal components method withvarimax rotation was used to extract 10 interpretable factors.Close analogues were obtained for the following Piers(1963) factors: Behavior, Physical Appearance and Attri -butes, Intellectual and School Status, and Anxiety. Thestudy found several other factors that diverged somewhatfrom the Piers model, including Object of Depreciation,Self-Significance, Personal Confidence, Self-Acceptance,and Excelling Academically.

Platten and Williams (1979, 1981) conducted twostudies that attempted to replicate the Piers-Harris factorstructure in ethnically mixed samples. Both samples includ-ed fourth- through sixth-grade students. The first sampleconsisted of 159 African American and Mexican Americanstudents, and the second sample included 193 Anglo-American, African American, and Mexican American stu-dents. These studies also featured a test-retest methodology:The factor stability in each sample was examined across a10-week retest interval. In both samples, a principal compo-nents solution with varimax rotation yielded 10 factors forboth pretest and posttest. However, only five of these factorshad sufficient commonalties on both administrations to war-rant joint interpretation. The five factors only partially repli-cated the Piers (1963) factor structure (e.g., in Platten andWillams, 1979, items representing physical appearance andschool status were combined on a single factor). The authorscautioned that the observed instability of the factor structurebetween consecutive administrations could limit researchers’ability to measure change in self-concept. Without an appro-priate control group, it might not be possible to partial outchange due to treatment effects versus change due to mea-surement error and other factors. Platten and Williams(1981) also proposed that a unidimensional model mightbetter represent the Piers-Harris factor structure than themultidimensional model proposed by Piers (1963).

The utility of a unidimensional model, in addition tothe effects of positive versus negative item phrasing, wereinvestigated in a confirmatory factor analysis reported byBenson and Rentsch (1988). These investigators used a sam-ple of 885 Anglo-American, African American, andHispanic American children in Grades 3 through 6. Theytested various structural-equation models for the Piers-Harris. The best model-data fit was observed for a three-factor model consisting of one general self-concept factorplus two item-phrasing factors. Thus, additional support wasprovided for the relatively greater validity of Piers-HarrisTotal score as compared to a multidimensional model of thescale (i.e., Piers’s six-factor model, 1963). (Note: similaritem-phrasing effects were suggested by the item compositionof the factors in the Piers-Harris 2 factor analysis describedpreviously.)

Benson and Rentsch (1988) also compared model fitbetween the full 80-item version of the Piers-Harris and a

shortened 65-item version obtained by removing 15 itemsnot scored on any cluster scale. They found nearly identicalmodel-fit statistics and internal consistency values for thetwo versions, and concluded that these 15 items “did not real-ly enhance the explanatory power” of the Piers-Harris.These findings provided part of the conceptual support forthe revision process that led to the 60-item Piers-Harris 2.

Williams, Apenahier, and Haynes (1987) reported afactor analysis of the Piers-Harris with a sample of 486 pre-dominantly African American gifted students. The studyemployed a principal components method with varimax ro-tation. Six interpretable factors were extracted and labeledFriendliness and Popularity, Physical Appearance, Satis -faction, Anxiety, Behavior, and School and Family. Althoughthese labels are similar to the Piers-Harris cluster scales, theauthors reported that the order of extraction and item contentof the factors differed from those identified by Piers in heroriginal factor analysis. Williams et al. suggested that theracial composition and intellectual giftedness of the samplemight have accounted for the discrepancies between theirfactor solution and the Piers factor structure.

Hattie (1992) described a restricted factor analysis ona sample of 367 students, in which six factors were hypoth-esized and items were constrained to load only on factorsthey were assigned to in the Piers-Harris cluster-scale scor-ing system. Several goodness-of-fit indexes indicated ade-quate fit of these data to the model. The item-loadingconstraints were then lifted and the resulting factor solutionwas rotated to a target matrix. The resulting item loadingsindicated support for the first five cluster scale labels, butsuggested that the HAP scale is actually bifactorial. This lat-ter finding is similar to the Piers-Harris 2 factor analysis.

Moller and Schnurr (1995) factor-analyzed the Piers-Harris in a sample of 140 adolescents (aged 10 to 17) whowere hospitalized at an inpatient psychiatric facility. A prin-cipal components analysis with oblique rotation was used.The analysis replicated all of the original factors except forthe Physical Appearance and Attributes cluster. The 14items composing this scale were subjected to an additionalanalysis using the same method. This analysis yielded twofactors, labeled Physical Appearance and Personality Attri -butes, accounting for 63% of the variance. Thus, Moller andSchnurr found that the PHY cluster scale was actually bi-factorial, which is consistent with the results of the Piers-Harris 2 factor analysis.

The most recent published factor analysis of the orig-inal Piers-Harris (Hur, McGue, and Iacono, 1998) used asample of 407 monozygotic and dizygotic twin pairs, aged11 to 12 years. A principle components analysis with vari-max rotation yielded a six-factor solution, with factorsroughly parallel to the original Piers-Harris cluster scalesbut with a few differences in item assignments. The authorsreported that their factor-analytically derived scales yieldedinternal consistency reliability coefficients (alphas) rangingfrom .66 to .83, with a mean of .75.

In summary, a number of factor-analytic studieshave investigated the underlying dimensions of the original

Chapter 5 Technical Properties 61

Piers-Harris. Six of the studies discussed in this section (Huret al., 1998; Michael et al., 1975; Moller & Schnurr, 1995;Piers, 1973; Rich et al., 1979; Wolf et al., 1982) have repli-cated many or all of the factors identified in the originalanalysis (Piers, 1963). The credibility of this factor solutionis enhanced by studies suggesting that roughly these samefactors also replicate across different ethnic minorities (Wolfet al., 1982); for various age groups (Michael et al., 1975);and even in samples of students with mental retardation(Rich et al., 1979) and adolescent psychiatric patients (Moller& Schnurr, 1995).

However, many of the studies either failed to replicateall six original factors or identified new factors. Two studies(Hattie, 1992; Moller & Schnurr, 1995) converge with thePiers-Harris 2 factor analysis to indicate that the PhysicalAppearances and Attributes scale and the Happiness andSatisfaction scale may be bifactorial. In contrast, two addi-tional studies (Benson & Rentsch, 1988; Platten & Williams,1981) favor a unidimensional factor structure over the multi-dimensional model established by Piers’s original factoranalysis. Although this conflicting evidence indicates a needfor further analysis of the structural characteristics of thePiers-Harris scales, most of the findings reviewed here dosupport the scales—with their traditional item assignments—as representing separate but interrelated aspects of self-concept. Although several authors have suggested re-assignment of items to different scales, the data they reportsimply are not strong or consistent enough to justify such amajor revision, which would have serious consequences forthe backwards-compatibility of the revised scales with theexisting Piers-Harris research database.

Convergent validity. Investigations of the psycho-metric properties of the original Piers-Harris have included

studies of convergent validity, which refers to the strengthof correlation between the Piers-Harris and measures of the-oretically similar constructs.

Correlations with other self-concept measures.Studies examining the relationship between the originalPiers-Harris Total score and other self-concept instrumentsare summarized in Table 26. Most of the correlation co-efficients are in the moderate to high range, thus supportingthe construct validity of the Piers-Harris. Some of the weakerrelationships may result in part from the effects of ageand/or differences in assessment format. For example, in theBolea, Felker, and Barnes (1971) study, the authors used ameasure of self-concept specifically designed for youngerchildren (Pictorial Self-Concept Scale). Some of the chil-dren in their sample, which ranged from kindergarten tofourth-grade students, were below the recommended agerange for the Piers-Harris, which may partially account forthe moderate validity coefficient (r = .42).

At the other end of the age range, a study using a sam-ple of 10th graders (Yonker, Blixt, & Dinero, 1974) may re-flect differences in questionnaire format. These authors foundmoderate correlations (r = .42 for males, r = .40 for females)between Piers-Harris Total score and the adult form of theBills Index of Adjustment and Values (Bills, 1973). As wouldbe expected, studies using the 1967 Coopersmith Self-EsteemInventory (Johnson, Redfield, Millerand, & Simpson, 1983;Schauer, 1975), which resembles the Piers-Harris in formatand age range, report higher validity coefficients (rs = .63 and.85, respectively). Differ ences between the paper-and-penciland computer-administered formats of the Piers-Harris wereexamined by Simola and Holden (1992). Both versions corre-lated strongly (rs = .68 to .73) with the Harter Self-PerceptionProfile for Children (Harter, 1985), which uses a forced-

62 Technical Guide

Table 26Studies Reporting Correlations Between Piers-Harris Total Score and Other Self-Concept Measures

Study Sample Age or grade Sex n Measure r

Bolea et al. (1971) Normal Grades K–4 Both 63 Pictorial Self-Concept Scale .42

Johnson et al. (1983) Normal Grade 5 Both 105 Coopersmith Self-Esteem Inventory .63

Karnes and Wherry (1982) Normal Grades 4–6 Not reported 96 Children’s Personality Questionnaire .34–.73

(O Factor)

Lewis et al. (1995) Gifted Grades 4–12 Both 368 Reflections of Self by Youth Scale .43

Mayer (1965) Special Education 12–16 years Not reported 98 Lipsett Children’s Self-Concept Scale .68

Parish and Taylor (1978a) Normal Grades 3, 6 Both 75 Personal Attribute Inventory for Children .67

Parish and Taylor (1978b) Normal Grades 3–8 Both 390 Personal Attribute Inventory for Children .32

Grades 5–8 Both 297 Nonsexist Personal Attribute Inventory .49

for Children

Schauer (1975) Normal Grades 5–6 Both 215 Coopersmith Self-Esteem Inventory .85

Simola and Holden (1992) Normal 8–17 years Both 206 Harter Self-Perception Profile for Children .68–.73

Yonker et al. (1974) Normal Grade 10 Male 100 Tennessee Self-Concept Scale .51

Female 108 Tennessee Self-Concept Scale .61

Male 100 Bills Index of Adjustment and Values .42

Female 108 Bills Index of Adjustment and Values .40

choice format in which children must decide between twoself-descriptive statements for each item.

The lowest correlation coefficient in Table 26 (r = .32)was reported in a study comparing the Piers-Harris to thePersonal Attribute Inventory for Children (Parish & Taylor,1978a, 1978b). The study used a sample of 390 thirdthrough eighth graders. The authors reported that the overallcorrelation coefficient was influenced by the weak relation-ship between the two measures (r = –.08) among the eighthgraders. Parish and Taylor suggested that the eighth-gradedata might have resulted in part from differences in the con-structs being measured by the two instruments, or from therelative emotional instability of this age group. In a study bythese same authors using younger children (Parish & Taylor,1978a), a much more robust association was observed be-tween the two measures (r = .67).

A more detailed perspective on convergent validitycan be gained by examining the relationship between thePiers-Harris and the second edition of the Tennessee Self-Concept Scale (TSCS:2, Fitts & Warren, 1996). The TSCS:2yields subscale scores that are similar to those of the Piers-Harris cluster scales. In addition, the TSCS:2 is available ina child form that has a recommended age range (7 to 14years) similar to that of the Piers-Harris. The primary differ-ence between the two measures is that the TSCS:2 uses a 5-point, Likert-type response scale, and the Piers-Harrisemploys a yes/no answer format. Table 27 presents inter-scale correlations between the two measures, with coeffi-cients in bold type representing relationships betweentheo retically similar scales (e.g., Piers-Harris Popularity and

TSCS:2 Social Self-Concept). The table indicates moderateto strong correlations between similar scales, supporting theconstruct validity of the Piers-Harris. The correlation be-tween the total scores (r = .66) is similar in magnitude to thevalidity coefficient reported in an earlier comparison of thetwo measures (Yonker et al., 1974).

Another recent study (Huebner, 1994) examined therelationship between the Piers-Harris scales and the Stu -dents’ Life Satisfaction Scale (SLSS; Huebner, 1991), ameasure thought to be conceptually similar to the Piers-Harris Happiness and Satisfaction scale. The study used asample of 235 nonclinical students in Grades 5 through 8.As expected, the author found that, among the Piers-Harrisscales, the Happiness and Satisfaction scale was the moststrongly related to the SLSS (r = .57). Correlations with theother Piers-Harris scales ranged from .29 to .37.

Correlations with other behavioral and personalitymeasures. Numerous investigators have examined the rela-tionship between the original Piers-Harris and other mea-sures of personality and behavioral functioning. Table 28summarizes several of these studies. Of particular interestare the studies on measures of emotional problems andpsychopathology. This research has supported the common-sense notion that positive self-concept should be inverselycorrelated with psychological and behavioral difficulties.Table 28 shows that the Piers-Harris Total score exhibits theexpected negative relationships with

1. SRA Junior Inventory, Big Problems subscale(Remmers & Bauernfeind, 1957; as cited in Cox,1996)

Chapter 5 Technical Properties 63

Table 27Correlations Between Piers-Harris Scores and Tennessee Self-Concept Scale 2 (TSCS:2) Scores

Piers-Harris scale

Intellectual Physical Happinessand Appearance andSchool and Satis-

TSCS:2 Scorea Behavior Status Attributes Anxiety Popularity faction Total

Summary ScoresTotal Self-Concept .61 .53 .44 .50 .38 .48 .66Conflict –.44 –.44 –.15 –.32 –.22 –.17 –.40

Self-Concept ScalesPhysical .48 .44 .44 .56 .36 .47 .62Moral .57 .42 .22 .24 .14 .27 .43Personal .50 .50 .45 .44 .35 .50 .63Family .53 .36 .19 .35 .21 .32 .44Social .43 .46 .44 .42 .41 .40 .58Academic/Work .46 .59 .44 .38 .36 .33 .56

Supplementary ScoresIdentity .52 .49 .52 .56 .46 .45 .66Satisfaction .56 .51 .49 .54 .38 .48 .64Behavior .62 .41 .29 .53 .32 .34 .58

Note. N = 224, ages 7 to 14. Bold type indicates theoretically expected correlation. aThe Child form was used.

2. Children’s Manifest Anxiety Scale (Castaneda,McCandless, & Palermo, 1956; as cited in Millen,1966)

3. Junior Eysenck Personality Questionnaire, Neu -roti cism and Psychoticism subscales (Eysenck,1970; as cited in Chapman, 1973, and Tavormina,1975)

4. Children’s Depression Inventory (Kovacs, 1981;as cited in Saylor et al., 1984)

Another study (Richman, Brown, & Clark, 1984) comparedthe Piers-Harris cluster scales to the Depression AdjectiveChecklist (Lubin, 1967) and the Psychological ScreeningInventory (Lanyon, 1973) in a sample of 195 11th- and 12th-grade students. The Piers-Harris Behavior and Anxiety clusterscales demonstrated strong negative relationships with thesetwo measures of psychological distress, with the strongest as-

sociations found in females of high socioeconomic status.Several researchers have studied the relationship be-

tween locus of control and self-esteem. Tavormina’s (1975)finding of a moderate positive correlation between Piers-Harris Total score and a locus of control scale (Nowicki &Strickland, 1973) is typical of this research. With respect tothe Intellectual Achievement Responsibility Questionnaire(Crandall, Katkovsky, & Crandall, 1965), correlations withthe Piers-Harris are generally positive for acceptance of re-sponsibility for success, but much less consistent for accep-tance of responsibility for failure (Felker & Thomas, 1971;Piers, 1977). Another study found a moderate relationship(r = .36) between the Piers-Harris Total score and a measureof children’s perceptions of their ability to be persuasivewhen interacting with their peers (Wheeler & Ladd, 1982).

The developmental aspects of self-concept have alsoreceived attention from scholars. It is usually assumedthat parental child-rearing attitudes and practices play an

64 Technical Guide

Table 28Studies Reporting Correlations Between the Piers-Harris and Other Behavioral and Personality Measures

Study Sample Age or grade Sex n Measure r a

Austin and Huberty (1993) Chronically ill 8–12 years Both 269 Child Attitude Toward Illness Scale .48**

Chapman (1973) Normal 10–11 years Both 455 Junior Eysenck Personality Questionnaire

Psychoticism scale –.27**

Extraversion scale .41**

Neuroticism scale –.34**

Cox (1966) Normal Grades 6–9 Both 97 SRA Junior Inventory

Big Problems scale –.64**

Health Problems scale –.48**

Felker and Thomas (1971) Normal Grade 4 Both 66 Intellectual Achievement Responsibility

Questionnaire (IAR) Success scale .32*

Millen (1966) Normal Grade 4 Both 275 Children’s Manifest Anxiety Scale (CMAS) –.59**

Children’s Social Desirability Scale .42**

Grade 6 Both 265 Children’s Manifest Anxiety Scale –.65**

Children’s Social Desirability Scale .45**

Grade 8 Both 231 Children’s Manifest Anxiety Scale –.61**

Children’s Social Desirability Scale .39**

Grade 10 Both 221 Children’s Manifest Anxiety Scale –.54**

Children’s Social Desirability Scale .25**

Grade 12 Both 191 Children’s Manifest Anxiety Scale –.69**

Children’s Social Desirability Scale .34**

Piers (1977) Normal Grades 6, 10 Both 297 IAR (Success) .35**

IAR (Failure) –.04

IAR (Total) .19**

Saylor et al. (1984) Psychiatric 7–11 years Both 24 Children’s Depression Inventory –.64**

inpatients

Tavormina (1975) Chronically ill 6–18 years Both 94 Junior Eysenck Personality Questionnaire

Neuroticism scale –.47**

Extraversion scale .49**

Nowicki-Strickland Locus of Control .35**

Wheeler and Ladd (1982) Normal Grades 3–5 Both 127 Children’s Self-Efficacy for Peer

Interaction Scale .36**aCorrelation with Piers-Harris Total score.

*p < .05. **p < .01.

important role in the development of positive or negativeself-attitudes. Several studies have examined parents’ state-ments about, or children’s perceptions of, these attitudes andpractices. For example, Cox (1966) found that the Piers-Harris Total score was positively associated with a child’sperception of his or her parents as loving (r = .56), and neg-atively correlated with parental disagreement over child-rearing practices (r = –.24). Piers (1972) comparedclinic-referred children to control children on parental pre-diction of children’s responses on the Piers-Harris. Parentsof clinic-referred children tended to underestimate their chil-dren’s self-esteem, and the reverse was true for the controlgroup. Although accuracy of prediction was similar betweenthe two groups, parents of clinic-referred children showedbetter agreement with their children on negative responses,and the control group parents displayed better agreementwith their children on positive responses. These findingssuggest not only the theoretically expected link betweenparental attitudes and children’s self-esteem, but also thelink between these attitudes and the parental perception thatchildren need or do not need professional help.

Correlations with parent, teacher, and peer ratings.Research examining the relationship between children’sself-ratings on the original Piers-Harris and parent, teacher,and peer ratings has been sparse, and much of what doesexist suffers from methodological weaknesses. For example,a study of the validity of the Achenbach Child BehaviorChecklist (Cohen, Gotlieb, Kershner, &Wehrspann, 1985)compared the Achenbach to a “short form” of the Piers-Harris, but did not provide any details on how the short formwas constructed. One hundred and four children (aged 6 to11) referred to a clinic for behavior problems were classi-fied as “Internalizers” or “Externalizers” based on their par-ents’ Achenbach ratings. These groups did not differ on aself-esteem score derived from the Piers-Harris short form.Another study (Waldron, Saphire, & Rosenblum, 1987) cor-related Piers-Harris scores with teacher ratings on theConners Teacher Rating Scale (Conners, 1969). This studyused a relatively small sample of 24 children who had beenidentified as having learning disabilities or as being intellec-tually gifted. Significant negative correlations (rs = –.49 to–.68) were reported among the Piers-Harris cluster scalesand the Conners Hyperactivity and Asocial Scales, suggest-ing that children with lower self-esteem were exhibiting ob-servable behavior problems in these areas.

Other research on informant ratings and the Piers-Harris has been in the form of unpublished studies, disser-tations, and theses. In samples of 111 fourth graders and 113sixth graders, Piers (1965) obtained correlations rangingfrom .06 to .49 between the Piers-Harris Total score andpeer and teacher ratings of self-concept. Peer ratings gener-ally exhibited stronger relationships with the Piers-Harristhan did teacher ratings. Cox (1966) studied children inGrades 6 through 9 from 97 families and found significantcorrelations of socially effective behavior between the Piers-Harris Total score and social-skills ratings from teachers(r = .43) and peers (r = .31). In a study of 75 third and fourth

graders, 50 of whom had been diagnosed with articulationproblems, Querry (1970) found that teacher ratings of self-concept correlated significantly with the Piers-Harris Totalscore for children without articulation problems (r = .54),but significantly for children with mild and moderate articu-lation problems (r = .26 and .02, respectively). The upshotof these various studies is that children’s self-concept rat-ings are often associated in expected ways with informantratings of theoretically related behaviors, but the strength ofthese associations varies widely with the particular groupunder study and the quality of the research method.

Multitrait-multimethod studies. The multitrait-multimethod research design (Campbell & Fiske, 1959) isgenerally viewed as a highly effective technique for evaluat-ing construct validity. In this methodology, the test of interestis compared with other measures of the same construct (e.g.,self-concept), and with measures of theoretically differentconstructs (e.g., academic achievement). Ideally, the variousconstructs are measured by at least two different assessmentmethods (e.g., interview vs. self-report questionnaire). Theaim of the multitrait-multimethod design is to demonstrateboth convergent and discriminant validity. Convergent valid-ity, as described previously, requires significant correlationsbetween various measures of self-concept. Discriminantvalidity, on the other hand, requires substantially lowercorrelations between measures of self-concept and mea-sures of other traits or constructs. The research literatureon the original Piers-Harris includes several multitrait-multimethod studies.

Winne, Marx, and Taylor (1977) examined the constructvalidity of the Piers-Harris by comparing it to the Sears Self-Concept Inventory (Sears, 1966) and the Gordon How I SeeMyself Scale (Gordon, 1968). The sample consisted of 103third through sixth graders of mixed ethnicity. Subtests ofthe three measures were grouped into three major categories(physical, social, and academic self-concept). Strong conver-gent validity was demonstrated for the physical appearancesubtests, some of the social interaction subtests (includingthe Piers-Harris Popularity scale), and the academic subtests.The evidence for discriminant validity was much weaker forthe three categories. This may be due in part to the lack oftrue heterogeneity of the assessment method. All three in-struments are self-report, although the authors expected thedifferences in response formats and comparison standards tooffset this fundamental similarity.

This study was replicated on a sample of 488 fourththrough sixth graders (Marx & Winne, 1978). Once again,the physical, social, and academic subscales showed goodconvergent validity. Discriminant validity was improvedfrom the first study, but was still inadequate to determine,for example, that these measures could reliably distinguishsocial self-concept from other facets of self-concept. Be -cause of the questionable discriminant validity, the authorsconcluded that both studies supported a unidimensionalmodel of self-concept.

Shavelson and Bolus (1982) reported a multitrait-multimethod study using a relatively small sample of 99

Chapter 5 Technical Properties 65

junior high school students. The two self-concept scales, thePiers-Harris and the Tennessee Self-Concept Scale (Fitts,1965) were conceptualized as measures of general self-concept. The average correlation between the two measures(r = .77) provided good evidence of convergent validity.Several measures of different constructs (e.g., perceptionof own abilities) were included in the study to complete themultitrait-multimethod matrix. Correlation coefficientsbetween general self-concept and these other measuresranged from –.02 to .38, providing evidence of acceptablelevels of discriminant validity.

A study by Franklin, Duley, Rousseau, and Sabers(1981) compared the Piers-Harris to the Coopersmith Self-Esteem Inventory in a sample consisting of 248 fourth-gradeand 321 seventh-grade students. The two measures corre-lated at r = .78, which established convergent validity. Dis -criminant validity was assessed by correlating self-conceptscores with variables representing academic achievement,socioeconomic status, special-education placement, ethnicity,grade placement, gender, and age. Multiple correlation co-efficients with these conceptually distinct variables did notexceed .25, which provides evidence of discriminant validity.Only academic achievement correlated significantly withself-concept scores (r = .18 for fourth graders; r = .22 forseventh graders). However, these relationships accounted forless than 5% of the variance in self-concept scores.

A more recent multitrait-multimethod study of self-concept instruments was reported by Marsh (1990). Thestudy used a sample of 290 fifth graders from Sydney,Australia. Three self-concept scales were examined: thePiers-Harris, the Self-Description Questionnaire (SDQ;Marsh, 1988), and the Perceived Competence Scale forChildren (PCS; Harter, 1979). As with earlier studies, sup-port for convergent validity was strong—correlation coeffi-cients among the three measures ranged from .45 to .69,with a mean of .61. However, the discriminant-validity evi-dence for the Piers-Harris was equivocal. For example, thecorrelation between the physical self-concept scales of thePiers-Harris and PCS (r = .45, same trait, different method)was lower than the correlation between the Piers-HarrisPhysical Appearance and Attributes scale and the PCSSocial scale (r = .56, different trait, different method). Thisrelationship is the opposite of what is needed to establishdiscriminant validity. Marsh concluded that these analysesprovided only partial support for the multidimensional na-ture of the Piers-Harris.

In summary, the multitrait-multimethod studies re-viewed here all uphold the convergent validity of the originalPiers-Harris, which is consistent with the other studies in-volving multiple measures of self-concept that were describedearlier in this chapter. In contrast, several of the multitrait-multimethod studies call into question whether the clusterscales of the original Piers-Harris can reliably discriminateamong different facets of self-concept. These findings mirrorconcerns expressed by Piers (1984) in her overview of thefactor-analytic research on the original Piers-Harris. Thecurrent analyses of the structural characteristics of the Piers-

Harris 2 (see earlier sections in this chapter) suggest that thedomain scales measure overlapping constructs, in part be-cause of shared item content among the scales. Nevertheless,the Piers-Harris 2 factor analysis yielded a six-factor solu-tion with item loadings roughly parallel to the item compo-sition of the six domain scales. This suggests that, despitethe thread of uncertainty running through the Piers-Harrisliterature, the domain scales do in fact provide useful infor-mation about empirically separable aspects of self-concept.

Validity evidence from intervention studies.Numerous investigators have used the original Piers-Harristo evaluate the outcome of psychological and educational in-terventions. These studies typically assume that such inter-ventions (e.g., psychotherapy) should have a positive impacton children’s self-concept. The construct validity of thePiers-Harris is supported if it can detect this hypothesizedenhancement in self-regard. The following section providesa review of selected intervention research, focusing on stud-ies with certain methodological strengths (e.g., large samplesize, use of a control group in an experimental design). Amore complete listing of outcome research is presented inAppendix A.

Therapeutic interventions. Children with acting-outdifficulties or externalizing behavioral disorders have re-ceived a great deal of attention from psychotherapy re-searchers. For example, Kendall and Braswell (1982)studied interventions aimed at improving the self-control ofchildren with behavioral problems. In the study, 27 children(aged 8 to 10) were randomly assigned to one of three treat-ment conditions: an attention control condition; a behavioraltreatment involving modeling and contingencies; and a cog-nitive-behavioral treatment, which added self-instructionaltraining to the behavioral interventions. Results indicatedthat the cognitive-behavioral group showed improvementsin self-concept and teacher ratings of self-control. Thesegains were maintained after a 10-week follow-up period, butdid not persist 1 year later.

Several studies have used the Piers-Harris to evaluateresidential treatment programs. Abidin and Seltzer (1981)investigated the effects of a residential program for 70 chil-dren with special needs (35 with learning disabilities, 8 withemotional disturbances, and 27 with learning disabilities andemotional disturbances). Pretest and posttest comparison re-vealed that the children’s self-concepts were significantlyhigher by the close of the academic term. Groves andKahalas (1976) reported that a residential-camp interventionimproved self-concept scores for teenagers whose baselinePiers-Harris scores were in the low to medium range. Therewas no improvement noted for teenagers who began thatstudy with high self-concept scores.

Social-skills interventions have been of interest to re-searchers, both for treating behavioral disorders and for im-proving the functioning of normal children. One study(Waksman, 1984a, 1984b) compared 30 adolescents who hadreceived a brief assertiveness-training intervention to acontrol group of 28 adolescents. The groups did not differin Piers-Harris scores taken at baseline. Following the

66 Technical Guide

intervention, the experimental group evidenced signifi-cant improvement in self-concept score, and these gainswere maintained at a 1-month follow-up assessment. Theeffectiveness of a social skills awareness program wasconfirmed in an experimental study (Wanat, 1983) involv-ing 30 high school students identified as having learning disabilities. Following the intervention, the Piers-HarrisTotal score was found to have improved significantly for theexperimental group, but not for the control group.

Cameron (1999) examined self-concept in obese chil-dren enrolled in a weight-management program. Of those inthe study, 49 obese children (aged 10 to 15 years) receivedthe intervention, and their Piers-Harris scores were com-pared with a control group of 60 obese children who hadnever participated in a weight-management program. Re -sults indicated that children in the experimental groupshowed significant declines in self-concept, particularly onthe Physical Appearance and Attributes cluster scale. Thisdecline may have occurred because the intervention forcedchildren to focus on undesirable aspects of their appearance.

The widespread use of stimulant medication for treat-ing Attention-Deficit/Hyperactivity Disorder (ADHD) hasled to interest in how the medication affects self-esteem.Frankel et al. (1999) compared Piers-Harris scores of 56children medicated for ADHD with those of a control groupof children who were not medicated. Results showed a dose-dependent relationship between stimulants and self-concept,with higher doses of medication associated with higherPiers-Harris scores. The Behavior, Intellectual and SchoolStatus, and Popularity cluster scores were significantly high-er in medicated children than in nonmedicated children. Aprior experimental study (Fehlings, Roberts, Humphries, &Dawe, 1991) determined that cognitive-behavioral therapywith ADHD children improves parents’ perceptions of chil-dren’s behavioral functioning, as well as improving chil-dren’s Piers-Harris scores. Werry, Aman, and Diamond(1980) studied the effects of imipramine on cognitive func-tion, self-concept, and other psychosocial variables in hyper-active children. A comparison group consisted of hyper activechildren being treated with methylphenidate. Imipramine ap-peared to improve learning, motor performance, and socialbehavior, but had only a minor effect on self-concept.

Several studies have examined interventions for im-proving self-concept in children of divorce. Burroughs,Wagner, and Johnson (1997) compared two interventions fortreatment of children of divorce: a board-game therapybased on coping-skills training and education about divorce,and a conventional form of play therapy. Results showedthat the children in both treatment conditions improved frompreintervention baseline on several measures of psychologicaladjustment; there was, however, no significant increase in Piers-Harris scores. In another study of treatments forchildren of divorce (Burke & Van de Streek, 1989), 39fourth to sixth graders from divorced single-parent homeswere randomly assigned to receive either eight sessions ofgroup counseling or a no-treatment control condition. Thosestudents in the experimental condition demonstrated

increases from baseline in their Piers-Harris scores.A primary prevention program for children of divorce

was investigated by Stolberg and Garrison (1985). In thestudy, 82 mother-child pairs were assigned to one of fourtreatment conditions: child support group only; single-parent support group only; both child and single-parentgroups; or no treatment. Results suggested that youngstersin the first category (child support group only) showed thegreatest increase in Piers-Harris scores.

Other intervention studies have documented expectedgains in self-concept for a variety of patient groups. Regan,Banks, and Beran (1993) examined the impact of a thera-peutic recreation program in 67 epileptic children. Resultsshowed that the intervention produced significant improve-ment in the children’s Piers-Harris scores, as well as in-creased acceptance of epilepsy. Rust and Troupe (1991)con ducted a study that showed that 6 months of group psycho-therapy had a positive impact on the self-concept scores of25 sexually abused girls. Lenkowsky, Barowsky, Dayboch,Puccio, et al. (1987) studied the effects of bibliotherapy onthe self-concepts of 96 adolescents with learning disabilitiesand emotional disturbances. Participants were assigned tofour reading groups, two of which focused on reading mater-ials related to the daily struggles of these students. The ado-lescents in the bibliotherapy condition demonstrated an in-crease in Piers-Harris scores.

Educational interventions. Studies in this area haveexamined classroom conditions (e.g., open vs. traditionalclasses, mainstreaming); training programs in specific skillareas; and tutoring. For example, Fitzpatrick (1975) studiedthe effects of a 16-week program of value-clarifying strat-egies on the self-concept of 547 seventh-grade students inparochial schools. Results indicated that the Piers-Harrisscores of the students who received this intervention increased more than the scores of a control group. Stevens and Pihl (1982) investigated a coping-skills training programfor teaching students to deal more adaptively with school-related stress. Seventh-grade students identified as likely tofail (n = 48) were randomly assigned to the coping-skillstraining program, a standard school-counseling program, ora control group. Although students in the experimentalgroup demonstrated improved social problem-solving abili-ty, their Piers-Harris scores did not increase significantly.

Reynolds and Cooper (1995) examined the effective-ness of a drug-prevention program directed at elementary-school children. The intervention was designed to enhancechildren’s social skills, improve academic performance, andreduce behavior problems. Children identified as being atrisk for drug problems participated in the study, with 77 chil-dren receiving the intervention and 35 children serving asthe control group. The prevention program was noted tohave a positive effect on behavior problems, as well as beingassociated with increases in self-concept as measured by thePiers-Harris. Franklin et al. (1981) studied the sensitivity ofthe Piers-Harris to changes in self-concept resulting from atutoring intervention. In the study, 184 fourth-grade studentswere randomly assigned to experimental or control conditions.

Chapter 5 Technical Properties 67

Students in the experimental group received 4 to 6 months oftutoring from high school students. No treatment effects onself-concept were observed.

Group-counseling methods for “at-risk” ninth graderswere investigated by Page and Chandler (1994). In this study,36 students were randomly assigned to a no-interventioncontrol group, an activity-oriented self-concept-buildinggroup, and a discussion-oriented group. It was found that theactivity-oriented group had a positive impact on self-esteem,especially the Piers-Harris Happiness and Satisfaction scale,when compared with the control group. Eldridge, Witmer,Barcikowski, and Bauer (1977) examined a group-guidanceprogram for improving the self-concepts of EMR students.This intervention was compared with individual teacher-selected methods for enhancing self-esteem in a sample of211 8- to 12-year-old students. Results indicated no differ-ence between the group and individual methods in terms oftheir effects on students’ Piers-Harris scores.

Researchers have studied interventions that are de-signed to increase self-concept and creative thinking amongnormal children. For instance, Kolloff and Feldhusen (1984)investigated the effects of a creative enrichment programwith a sample of 392 elementary school students who wereidentified as potentially gifted. Participants in the experi-mental condition took part in a pullout program in whichthey met twice weekly with a trained resource teacher. Atthe conclusion of the study, the experimental and controlgroups were found to differ on measures of creative think-ing, but not on Piers-Harris scores. Flaherty (1992) exam-ined the effects of a multimodal creativity program on theself-concept and school performance of third-grade stu-dents. In the study, 23 children received the intervention and22 were assigned to the control group. The experimentalgroup showed significantly increased scores on the Piers-Harris and on a measure of cognitive creativity.

Stefanich, Wills, and Buss (1991) examined the effectsof interdisciplinary team teaching on students’ self-concept.These investigators surveyed principals and teachers of 18middle schools to determine the extent to which each schoolimplemented interdisciplinary team teaching. The Piers-Harris was administered to 1,943 students at these schools.Results demonstrated that increased use of this teaching approach was associated with higher self-concept scores.

Benson, A. Kornhaber, C. Kornhaber, LeChanu, et al.(1994) studied the impact of a relaxation-training programfor high school students. In the study, 50 students were ran-domly assigned either to a health curriculum based on relax-ation training, or a control curriculum. Results suggestedthat the experimental group showed an increase in both self-concept (as measured by the Piers-Harris) and internal locusof control. Kolloff and More (1989) investigated changes inself-concept of 508 gifted students who had participated inone of three summer residential programs. The three pro-grams were associated with increases in self-concept scores,except for sixth-grade boys, who showed a decline in self-concept ratings.

Sorsdahl and Sanche (1985) examined the effects of a

classroom meeting/preventative counseling intervention onthe self-concept of elementary school students. The partici-pants in the study were 91 fourth-grade students from fourclassrooms. Two of the classrooms held problem-solvingand general discussion meetings for 20 weeks (experimentalcondition), while the other two classrooms participated inspecial activity periods (control condition). At the conclu-sion of the study, the experimental and control groups didnot differ on Piers-Harris scores. However, the experimentalgroup showed a significant increase in self-concept scorefrom preintervention baseline.

Cognitive restructuring as a primary prevention tech-nique was evaluated by Baker, Thomas, and Munson (1983).In the study, 58 ninth-grade students who met in weekly cog-nitive restructuring groups were compared with 67 ninthgraders who met in weekly discussion groups. The cognitive-restructuring intervention did not have a significant effect onself-concept scores. The effects of reality-therapy methodson the on-task behavior and self-concept of 150 fourthgraders were evaluated by Shearn and Randolph (1978).Students were randomly assigned to treatment and controlgroups. It was found that Piers-Harris scores did not improvefor students who received the experimental intervention.

Parent and teacher interventions. Another group ofstudies has focused on whether interventions designed to im-prove parenting and teaching skills also affect the self-esteemof children. For example, Giannotti and Doyle (1982) evalu-ated an 8-week parent-effectiveness training program with theparents of 46 students with learning disabilities. At the con-clusion of the study, children of parents receiving the traininghad significantly higher self-concept scores than children ofparents in the control group. Nevertheless, the Piers-Harrisscores of the experimental group remained lower than thoseof the children of the parents in the control group.

Esters and Levant (1983) evaluated the effectivenessof two parent-training programs in enhancing the self-esteem of rural, low-achieving children. In the study, 33 par-ents of third and fourth graders were randomly assigned tothe “systematic training for effective parenting condition,”the “self-esteem method” condition, or a wait-list controlgroup. After 10 counseling sessions, children from bothtreatment groups demonstrated improved academic achieve-ment, but only children from the self-esteem method groupshowed a significant increase in Piers-Harris scores.

Ascione and Borg (1983) investigated a teacher-training program designed to improve the self-concepts ofmainstreamed students with disabilities. The experimentalgroup included 15 elementary-school teachers who receivedthe training and their 251 students, of whom 47 were ethnicminorities and 38 were EMR. The control group included 18 teachers and 464 students, with similar proportions ofminority and mainstreamed students. Students’ Piers-Harrisscores were compared before and after the teacher training.Results demonstrated that self-concept scores improved forthe students with mental retardation, but not for the other stu-dents. In an earlier study, Borg (1977) examined the effectsof two teacher-training modules (classroom management vs.

68 Technical Guide

self-concept) on a variety of outcomes in sixth-gradestudents. This study found that although both modulesproduced significant changes in numerous teacher skills andstudent behaviors, students’ Piers-Harris scores did not in-crease from prestudy baseline.

Summary. Although the majority of the treatment-outcome studies reviewed in this section found that psycho-logical interventions produced expected changes inself-concept scores, more than a trivial number did not. Manyof these negative results can be attributed to weaknesses in theinterventions themselves, such as inadequate duration oftreatment, or to methodological flaws, such as the lack of ap-propriate comparison groups. Even in studies that featuredsuccessful interventions, it should be remembered that changein self-concept might have occurred as a result of attentionalone (i.e., being singled out for a special program) rather thanfrom the unique features of the program itself. Taken together,however, these studies do appear to support the utility of thePiers-Harris as a dependent measure in a wide range of pro-gram-evaluation and treatment-outcome applications.

Criterion Validity of the Original Piers-HarrisA primary index of the criterion validity of the original

Piers-Harris has been its ability to distinguish between groupsthat are expected to differ in self-concept (e.g., nonclinicalchildren vs. children hospitalized for psychiatric problems).This section provides an overview of research in this area.

A first attempt at confirming self-concept differencesacross criterion groups was made during the initial Piers-Harris standardization process. The scale was administeredto 88 institutionalized females with mental retardation, witha mean age of 16.8 years and a mean IQ of 70. As predicted,these adolescents scored significantly lower on the Piers-Harris than nonclinical children of either the same chrono-logical age or the same mental age (Piers & Harris, 1964).However, the authors speculated that these lower scores mayhave arisen more from institutionalization, and the factorsthat made it necessary, than from mental retardation per se.In support of this idea, Gorlow, Butler, and Guthrie (1963)reported that institutionalized children with mental retard-ation manifest significantly more negative self-attitudes thannoninstitutionalized children with mental retardation.

Mannarino (1978) examined self-concept among 92sixth-grade boys who were classified according to theirinvolvement in a “chum” relationship, which was definedby friendship stability, quality of communication, sensitivi-ty to the friend’s needs and interests, and preference forspending time with the friend. In the sample, 30 boys metcriteria for having a chum relationship. A comparison groupof 30 boys was selected, matched on level of social accep-tance and intelligence, but with fewer close relationshipswith peers. As predicted, the boys involved in a chum rela-tionship had significantly higher Piers-Harris Total scoresthan the boys in the control group.

Kinard (1980) investigated differences in self-conceptbetween physically abused children and a matched group ofnonabused children. Each group included 30 children,

ranging in age from 5 to 12 years. The groups did not differon mean Piers-Harris Total score, but significant differencesin individual item responses were noted. Compared with thecontrol group, the abused children were more likely to re-spond in the direction of negative self-concept on sevenitems—the ones depicting themselves as being sad, un-popular, unhappy, disobedient; wanting their own way;doing many bad things; and believing their parents expectedtoo much of them. Subsequently, a principal componentsanalysis was performed on this sample, yielding six factorsthat approximated those reported by Piers (1963, 1973).Com parison of the abused and control groups on the factorscores indicated a marginally significant difference on a factorlabeled Conformity-Happiness.

Pandina and Schuele (1983) studied the ability of thePiers-Harris to predict differences in substance abuse in twosamples: a group of 1,960 junior high school and high schoolstudents, and a group of 224 adolescents who had been diag-nosed with alcohol and/or drug abuse. These subjects wereclassified into six ranked categories to quantify their level ofsubstance use. In both samples, Piers-Harris Total score andBehavior scale score differentiated between the extremehigh and low categories of substance abuse.

Guiton and Zachary (1984) investigated differences inself-concept between nonclinical and clinic-referred chil-dren. The nonclinical group was the sample used for stan-dardizing the original Piers-Harris cluster scales (n = 485,age range of 10 to 16). The clinic sample consisted of 97children with a variety of psychiatric problems and an aver-age age of 10.4 years. The nonclinical group achieved signif-icantly higher scores than the clinic group on Piers-HarrisTotal and five of the six cluster scales. Only the Behaviorscale did not discriminate between the two groups. The au-thors speculated that children in the clinic group might haveresponded defensively to items representing problematic be-haviors, thus obscuring the difference between the twogroups on the Behavior scale. The two groups were alsocompared on their responses to each of the Piers-Harrisitems. Table 29 displays items that discriminated between thetwo groups, with their associated effect sizes. Of the 12 itemswith the largest effect sizes, only one loads on the Behaviorscale, which lends credence to the authors’ hypothesis aboutwhy that scale did not differentiate the two groups.

Kashani, Strober, et al. (1988) examined whether thePiers-Harris could differentiate between nonclinical adoles-cents and those with a psychiatric diagnosis. These investi-gators studied a group of 150 adolescents, of whom 28 hadmet criteria for a psychiatric diagnosis by structured clinicalinterview. The diagnoses included depressive disorder, anx-iety dis orders, conduct disorders, and alcohol or drug abuse.The mean Piers-Harris Total score for the “well-adjusted”adolescents was almost 17 points higher than for the psychi-atric cases, which represents a large effect size. In contrast tothese findings, Kelly (1988) reported that mean Piers-Harrisscores from a group of adolescents with conduct disorder(n = 38) differed from the Piers-Harris normative sampleonly on the Physical Appearance and Attributes scale,

Chapter 5 Technical Properties 69

whereas the scores from a group of adolescents with emo-tional disturbances (n = 43) differed from the normativesample only on the Physical Appearance and Attributesscale and the Popularity scale.

Taken together, these studies suggest that certain cri-terion groups, including children with mental retardation,socially isolated children, abused children, and children andadolescents with various forms of psychopathology, showexpected deficits in self-concept when compared with chil-dren without these conditions or circumstances. These find-ings support the criterion validity of the Piers-Harris.

Summary and Directions forFuture Research

The research findings reviewed in this chapter indicatethat the Piers-Harris 2 is a reliable and valid instrument forthe measurement of children’s self-concept. In terms of re-liability, the Piers-Harris 2 standardization study demon-strates that the revised measure has excellent internalstability, and the measure’s test-retest reliability is upheld bynumerous studies of the original Piers-Harris. In terms ofvalidity, the Piers-Harris 2 exploratory factor analysis yielded

70 Technical Guide

Table 29Piers-Harris 2 Item Endorsement by Clinic-Referred and Nonclinical Children

% Endorsing item in direction ofpositive self-concept

Item Clinic-referreda Nonclinicalb Effect size (d)c

1. My classmates make fun of me. (1) 66 80 .35

2. I am a happy person. (2) 84 93 .33

3. It is hard for me to make friends. (3) 67 82 .39

4. I am often sad. (4) 56 70 .31

5. I am smart. (5) 58 71 .30

9. I am a leader in games and sports. (63) 27 40 .26

11. I am unpopular. (11) 54 73 .42

14. I cause trouble to my family. (14) 69 81 .28

16. I am an important member of my family. (17) 64 82 .44

18. I am good in my schoolwork. (21) 59 80 .52

20. I behave badly at home. (25) 76 85 .23

21. I am slow in finishing my schoolwork. (26) 48 75 .60

23. I am nervous. (28) 56 69 .28

25. In school I am a dreamer. (31) 58 68 .22

31. I like being the way I am. (39) 68 83 .38

34. I often volunteer in school. (42) 72 56 .32

35. I wish I were different. (43) 56 72 .34

37. I am among the last to be chosen for games and sports.d (46) 55 70 .32

40. I am unhappy. (50) 77 86 .24

41. I have many friends. (51) 69 83 .37

42. I am cheerful. (52) 77 87 .28

45. I get into a lot of fights. (56) 61 76 .36

46. I am popular with boys. (57) 41 62 .44

47. People pick on me. (58) 55 72 .38

48. My family is disappointed in me. (59) 70 88 .50

51. In games and sports, I watch instead of play. (65) 60 81 .53

53. I am easy to get along with. (67) 72 84 .32

54. I am popular with girls. (69) 44 64 .40

58. I think bad thoughts. (78) 84 69 .33

Note. Only items that discriminated at p < .05 are listed. Numbers in parentheses are original Piers-Harris item numbers. Bold rows indicate items forwhich clinic-referred children scored higher than nonclinical children.aN = 97. bN = 485. cd = (mean clinical – mean nonclinical)/pooled SD. By convention, magnitudes of d corresponding to small, medium, and largeeffect sizes are .20, .50, and .80, respectively (Cohen, 1969). dThe phrase “and sports” was added for Piers-Harris 2 revision.

a six-factor solution that supports the assignment of itemsto six domain scales designed to measure separate, but inter-related, aspects of self-concept. This factor structure wasconsistent with several earlier factor analyses of the originalPiers-Harris. Construct validity is further supported bystudies indicating that the original and revised instrumentsshow expected relationships with self-concept question-naires, as well as with measures of other personality and be-havioral characteristics. Finally, several studies support thecriterion validity of the Piers-Harris by showing that it candifferentiate between groups that would be expected to differin self-concept.

The authors of the Piers-Harris 2 encourage continu-ing study of the psychometric properties of the instrument.Future research might address the following topics:

1. Use of confirmatory factor analysis and structural-equation modeling techniques for further valida-tion of the multidimensional nature of theinstrument

2. Replication of factor analyses across differentage, ethnic, and gender groups

3. Further exploration of the interaction effects ofsuch moderator variables as ethnicity and socio -economic status

4. Examination of the stability of self-concept ingroups with low Piers-Harris 2 scores

5. Further research on the ability of young childrento understand the Piers-Harris 2 items and fur-ther analyses of the psychometric properties ofthe scale when used with younger children

Chapter 5 Technical Properties 71