Research in Higher Education, Vol. 38, No. 5, 1997

LOWER-DIISION CLASS SIZE AT U.S .POSTSECONDARY INSTITUTIONS

Steve Chatman

The 1992-93 National Survey of Postsecondary Faculty (NSOPF) provides a uniqueopportunity to examine class size and the institutional and academic factors associ-ated with class size differences. The institutional characteristics examined in thispaper include public or private control, Carnegie classification, and institutional size.The course academic characteristics considered include academic discipline as de-scribed by Biglan (1973a), student level, and principal method of instruction. Theresults support the formation of comparative distributions for class size comparisonand better understanding of institutional and course characteristics associated withclass size differences.

The goal of this paper is to offer comparative standards for class size to aidadministrators faced with reduced funding and increased scrutiny. With veryfew exceptions (i.e., Bloom, 1983; Chatman, 1995) there are no empiricallyderived comparative standards for postsecondary instructional production thatrecognize even the most simplistic distinctions of academic discipline andcourse level or type of course. This lack of valid comparative information se-verely limits the ability of administrators facing a public that demands morewhile offering less. In the absence of comparative class-size norms and aca-demic measures that might provide a direction for change and a foundation fordialogue, public demands are logically expressed in simple efficiency goals—requiring that we do more with the same resources or the same with less. Thissituation was generally described by Zemsky and Massy (1995) as the inevita-ble result of Bowen's law of revenue and expense in times of economic con-traction. If "an institution of higher education will get all the money it can andspend all that it gets" then "the best way to make colleges and universities lesscostly is to give them less money" (p. 44). In the absence of standards of good

Dr. Steve Chatman, Director of Analytical Studies, Planning and Budget, 104 University Hall,University of Missouri, Columbia, MO 65211. chatmans@ext.missouri.edu.

Paper presented at the 1996 Annual Forum of the Association for Institutional Research, Albu-querque.

615

036I-0365#7/1000-06IS$12.50/0 © 1997 Human Sciences Press, Inc.

616 CHATMAN

practice or even comparative standards, institutions have little defense againstreductionist efficiency goals that consider only production costs. Among alter-natives to reduce instructional production costs in isolation or in combinationare to increase the number of classes taught by faculty, to increase class sizeand adopt instructional methods appropriate to larger classes as necessary (es-pecially technologically based strategies), and to increase reliance on part-time,adjunct faculty, and teaching assistants. Whichever alternative is employed,comparative information about class size would inform the policy formationdiscussion.

Institutional researchers who would try to respond to efficiency concernsusing valid, externally recognized standards for class size are hamstrung. Stan-dards of good practice are not available. Instead, institutional researchers areforced to rely on crude student/faculty ratios, student credit-hour weightingschemes of assumed validity, decades-old state formula standards, or local his-torical practice. In the absence of published standards of good practice, institu-tional researchers might turn to comparative information, but even the disciplin-ary information shared by recent cooperative data exchanges has not directlysupported comparative class size analysts by type of course. These exchangesinclude the American Association of University's Data Exchange, ClemsonUniversity's National Teaching Load Data Exchange, and the University of Del-aware's National Study of Instructional Costs and Productivity by AcademicDiscipline.

This article does not address the relationship of class size and studentachievement, satisfaction of student or teacher, or change in student behavior orattitudes. These outcomes are critically important and must be weighted whendecisions about class size are made, but they are beyond the scope of thisarticle and have been the subject of much controversy. On one hand, publishedstudies have typically shown achievement to be independent of class size at thepostsecondary level in the range of most lower-division classes (Glass andSmith, 1979; Levin, 1988; Slavin, 1990; Williams et al., 1985). On the otherhand, both students and faculty are more satisfied in smaller classes (Glass etal., 1982; Siegfried and Fels, 1979, pp. 938-939), they are more likely to enrollin future courses in the discipline if in smaller classes (Maxwell and Lopus,1995), and smaller classes may produce better outcomes when the measure iscritical thinking or attitudinal change (McKeachie, 1980, pp. 26-27). Instead ofcontributing to this ongoing discussion, this article will be satisfied with estab-lishing comparative standards for class size.

Assuming that comparative standards would serve a useful function, howmight they be gathered? One solution would be to interview randomly selectedfaculty nationwide about the classes they teach and compile that informationaccording to factors demonstrated to be associated with class size. A project togather comparative national information about course sections taught would

LOWER-DIVISION CLASS SIZE 617

require Herculean effort, be expensive, and demand strong support from facultyand administrators across the country if a reasonable sample were to be identi-fied and interviewed. These characteristics would make such a study practicallyunlikely and, if the purpose were described as forming comparative standardsfor class sizes, politically unlikely as well. Fortunately, that information can begleaned unobtrusively from a project that has been completed for other pur-poses: the 1992-93 National Survey of Postsecondary Faculty. Because thesurvey gathered specific information about each course section the randomlyselected faculty taught, the extent to which institutional and course characteris-tics were associated with class size can be used to create comparative distribu-tions.

1992-93 NATIONAL SURVEY OF POSTSECONDARY FACULTY

The 1992-93 National Survey of Postsecondary Faculty was the second cy-cle of the U.S. Department of Education's National Center for Education Statis-tics (NCES) study of U.S. faculty at public and independent not-for-profit insti-tutions. Specifically, the 1987-88 and 1992-93 surveys were to provide anational profile of the professional background, responsibilities, workloads, sal-aries, benefits, and attitudes of postsecondary faculty. The first cycle was in1987-88 when information was gathered from a sample of over 480 institu-tions, 3,000 chairpersons, and 11,000 faculty. The 1992-93 cycle was limited toinstitutions and faculty but was expanded to include samples of 974 institutionsand 31,354 faculty. Of these, 817 institutions and 25,780 faculty participated forresponse rates of 85% and 87%, respectively.

A key difference between the 1987-88 and 1992-93 surveys was the criteriafor inclusion. Both the 1987-88 and 1992-93 surveys included faculty whoregularly had teaching assignments, but the 1992-93 survey also included fac-ulty and administrators with faculty positions who did not regularly teach andemployees who taught whether or not they were considered to be "faculty."Specifically excluded were those with instructional duties outside the U.S., ifnot on sabbatical, temporary replacements, faculty on unpaid leave, militarypersonnel teaching only ROTC courses, instructional personnel supplied by in-dependent contractors, and teaching assistants. Of these inclusions and exclu-sions, the one that most limits use of NSOPF 1992-93 in developing compara-tive class-size distributions for all instruction is the exclusion of graduateteaching assistants from the sample universe. In spite of this limitation, NSOPF1992-93 remains a valid source of class size information for classes taught byall faculty. This issue will be discussed in more detail in the "Results" section.

This study focuses on a series of data supplied by faculty in response toSection C of the questionnaire, Institutional Responsibilities and Workload,where faculty described the characteristics of up to five courses that they taught

618 CHATMAN

in the fall of 1992. The class characteristic items included academic discipline,credit-hour value, number of teaching assistants, students enrolled, primarylevel of students, and primary instructional method used. In all, faculty de-scribed 56,959 classes. While the class characteristic data were collected tosupport study of faculty workload, they also provide a comparative base for thestudy of class size.

METHODOLOGY

It is asserted that the exploratory nature of this study should justify reason-able leeway in analytical procedure to accomplish the goal of simplifying thevariable structure of the database. The first restriction was to limit analysis tolower-division courses. It was assumed that course level would have a largeexplanatory effect that could be controlled by restriction to a single level.Lower-division courses were selected because they were the more numerousacross institution types and were less likely than upper-division or graduatecourses to be affected by student demand for courses required in the major.Other restrictions were to eliminate courses with credit values less than one orgreater than five, to remove institutions without NCES identification codes, andto remove classes of one or fewer students. The credit value, NCES code, andsize restrictions were to remove exceptional or extreme cases that were likelyoutliers. With these restrictions, the distribution of class size was subjected to aseries of main effects general linear models to find categorical variables thatexplained differences observed in lower-division courses and to reduce thecomplexity of categorical variables. The general linear model was used to ac-complish the analysis of variance solution because of the large differences incluster sizes for some variables.

To simplify the data structure, the obviously skewed distribution of classenrollment size was converted to its natural logarithm to produce a symmetricdistribution before the main effects models were tested. This phase of the anal-ysis, main effects models testing, sought to eliminate irrelevant variables fromfurther study and to reduce the levels of important variables. Independent vari-ables considered included Carnegie classification, public or private control, in-stitutional size, and Biglan classification of the academic material of the coursealong the three dimensions: pure versus applied, hard paradigm versus softparadigm, and life versus nonlife or inanimate. The first two of these measures,Carnegie classification and public or private control, are straightforward. Thethird variable, institutional size, used the classification scheme employed byNCES of institutions according to undergraduate full-time-equivalent enroll-ment with breakpoints at 631, 1,485, 3,564, and 7,788. The last measure, ormore accurately the last series of measures, was Biglan academic classificationof the course. The classification scheme was asserted as necessary due to the

LOWER-DIVISION CLASS SIZE 619

specificity and associated small sample size at which course academic contentwas identified. The Biglan structure was selected over other disciplinary cate-gorizations, such as two-digit CIP clusters, because previous research hasshown it to be a valid system and its use would maintain large numbers ofclasses in all clusters.

Among the evidence validating Biglan's assertion of three inclusive dimen-sions are the following. There was Biglan's first reported research where hestated that the classification scheme helped to explain differences in social con-nectedness of faculty, commitment to teaching or research, scholarly output(number of journal articles, monographs, and technical reports published by thefaculty), quality of first positions after completing degrees, and number of dis-sertations sponsored (1973b). This first work has been followed by much val-idation by other authors supporting the typology as a reasonable system bywhich to cluster higher education academic content. For example, Biglan cate-gories differentiate disciplines by faculty salaries and staffing patterns (Muffoand Langston, 1981; Smart and McLaughlin, 1978), departmental goals andactivities (Smart and Elton, 1975), instructional objectives (Smart and Eth-ington, 1995), job satisfaction (Eison, 1976), professional development needs(Creswell, Seagren, and Henry, 1979), and research productivity and interest(Creswell and Bean, 1981). A good introduction to the Biglan typology and itsuses is available in Creswell and Roskens's 1981 article in The Review ofHigher Education.

The classes were categorized along Biglan dimensions by applying Biglan'sown placement of 35 academic departments (1973a) and Malaney's extensionto 114 graduate-degree programs (1986). The very few remaining cases noteasily placed by Biglan's or Malaney's examples were placed by their strongassociation with known instances or were dropped. (The classes dropped be-cause of academic field were mostly vocational and technical.) Malaney'splacement of academic programs by Biglan dimensions appears as Table 1.

RESULTS

As described in the "Methodology" section, general linear model solution toanalysis of variance was performed to determine whether any of the main ef-fects variables could be dropped from further study and whether the variableclass levels of the remaining main effects could be combined to simplify study.Seven main effects were examined: instructional method, institutional size, Car-negie classification, institutional control, and the three Biglan topology dimen-sions.

Simple descriptive statistics for the seven main effects appear in Table 2.Overall, the mean class size was 29. By instructional method, the largest classeswere lectures (32) and radio or TV courses (30). These two were followed by

620 CHATMAN

TABLE 1. Biglan -Type Classification of Graduate Academic Programs"

High-Paradigm DevelopmentPure/NonlifeAstronomyAtmospheric ScienceChemical PhysicsChemistryGeology and

MineralogyMathematicsPhysicsStatistics

Pure/LifeAnatomyAnimal ScienceBiochemistryBiophysicsBotanyEntomologyEnvironmental BiologyGeneticsInterdisciplinary

Biochemistry

Pure/NonlifeClassicsEast Asian Languages

and LiteratureEnglishGeographyGermanHistoryLiberal StudiesLinguisticsPhilosophyRomance Languages

and LiteratureSlavic Languages and

Literature

Pure/LifeMedical Microbiology,

and ImmunologyMicrobiologyMolecular, Cellular,

and DevelopmentalBiology

PhysiologyPlant PathologyPreventive MedicineVeterinary Preventive

MedicineZoology

Applied/NonlifeAeronautical and

AstronauticalEngineering

AgriculturalEngineering

Ceramic EngineeringChemical EngineeringCivil EngineeringComputer and

Information ScienceElectrical Engineering

Low-ParadigmApplied/Nonlife

Accounting andManagementInformation Systems

ArchitectureArtBusiness

AdministrationCity and Regional

PlanningCommunicationDanceEconomicsHistory of ArtJournalism

Applied/NonlifeEngineering MechanicsGeodetic ScienceIndustrial DesignIndustrial and Systems

EngineeringMechanical

EngineeringMetallurgical

EngineeringMining EngineeringNuclear EngineeringPathologyPharmacyWelding Engineering

Applied/LifeAgronomyBiomedical

EngineeringDairy ScienceDentistryFood Science and

NutritionHorticultureHuman Nutrition and

Food Management

DevelopmentApplied/Life

AgriculturalEconomics

Agricultural EducationAllied Medical

ProfessionsArt EducationEducational Policy and

LeadershipEducational Theory

and PracticeFamily Relations and

HumanDevelopment

Family ResourceManagement

Applied/LifeMedicineObstetrics and

GynecologyOpthhalmologyPediatricsPharmacologyPhysical MedicinePhysiological OpticsPoultry ScienceRadiologySpeech and Hearing

ScienceSurgeryVeterinary AnatomyVeterinary Clinical

ScienceVeterinary

PathobiologyVeterinary Physiology

and Pharmacology

LOWER-DIVISION CLASS SIZE 621

TABLE 1. (Continued)a

Pure/LifeAnthropologyBlack StudiesPolitical SciencePsychologySociology

Low-Paradigm DevelopmentA p p l i e d / N o n l i f e A p p l i e d / L i f e

Landscape Health, PhysicalArchitecture Education and

Music RecreationPhotography and Home Economics

Cinema EducationPublic Administration Hospital and HealthTextiles and Clothing ServicesTheatre Administration

Human ServicesEducation

Labor and HumanResources

Natural ResourcesNursingSocial WorkVocational Education

This table represents the classification of all 114 graduate academic degree programs at a large public researchuniversity in the midwest. Each program is classified along the three Biglan dimensions: high/low-paradigmdevelopment, nonlife/Iife area of study, and pure/applied research emphasis.Source: From "Differentiation in Graduate Education" by Gary D. Malaney, 1986, Research in Higher Education25(1), p. 88. Copyright 1986 by Agathon Press, Inc.

the other seven methods whose means differed by three students or less. Thepercentile distributions for these instructional methods clearly show great over-lap. In fact, the figures presented show only about a 25% shift in percentilesfrom the smallest to largest instructional methods. This large overlap in distri-butions is similar across all comparisons and obviously limits the ability ofstatistical treatments to find important explanatory variables.

For the other main effects variables the following was true: Classes weresmaller on average at smaller institutions; liberal arts institutions had smallerclasses than public two-year institutions and public two-year institutions hadsmaller classes than comprehensive, doctoral, and research institutions; classeswere smaller at private institutions; classes in applied disciplines were smallerthan in pure disciplines; inanimate disciplines had smaller classes than life sci-ences; and classes with hard and soft paradigms were the same in size. How-ever, these statements were based on analysis of the raw numbers of theseobviously skewed distributions. Analysis of variance demands the more sym-metric distributions that can be created by analyzing the distribution of naturallogarithms. The results of this nonlinear transformation of class size are inTable 3. Analysis of variance examination of mean differences for natural log-arithms shows that the observations made about the untransformed distributionswere generally true.

622 CHATMAN

TABLE 2. Descriptive Statistics for Lower-Division Class Sizes

Percentile

TotalInstructional Method

1. Role-Play,Simulations, or OtherPerformanceActivities2. Seminar3. Lab, Clinical, orProblem Solving4. Apprenticeship,Internships, FieldWork, or Field Trips5. CooperativeLearning Groups6. Group Projects7. Discussion Groupsor Class Presentation8. TV or RadioCourses9. Lecture

Instructional MethodCollapsed

Lab, Performance,and Seminar (1, 2, 3,4 above)Discussion (5, 6, 7, 8above)Lecture (9 above)

Institutional Size1-631632-1,4851,486-3,5643,565-7,788Above 7,788

Carnegie ClassificationLiberal ArtsOtherPublic Two-yearComprehensiveDoctoralResearch

Mean

29

2222

22

23

2424

25

3032

22

2432

2124263038

242626334054

SD

25

1816

18

22

1015

15

2627

18

1427

1517182238

191816254164

N

29,768

1,155903

3,256

163

758108

3,607

7819,740

5,477

4,55119,740

1,3884,0539,2189,3635,746

2,1932,110

16,7186,0031,5901,154

10%

12

89

10

6

1212

12

914

9

1214

610121515

91012141415

25%

18

1215

14

12

1815

18

1520

14

1820

1215172022

151518202020

50%

25

2020

20

20

2422

24

2527

20

2427

1821252530

222425262830

75%

34

2525

25

26

2728

28

3535

25

2835

2529303540

303130364260

909?

45

3535

35

40

3536

35

5550

35

3550

3537404560

4042405078

125

LOWER-DIVISION CLASS SIZE 623

TABLE 2. (Continued)

Percentile

PrivatePublicAppliedPureInanimateLifeHardSoft

Mean

2730263227343129

SD

2525232622302922

N

5,87123,89712,66717,10121,635

8,13310,04919,719

10%

1012101512131212

25%

1520152018201818

50%

2225222625282525

75%

3035303530403433

90%

4145404740554845

Class sizes at smaller institutions were smaller and the relationship betweeninstitutional size and class size was positive and monotonic. Classes weresmaller at private institutions, in applied and inanimate disciplines, and disci-plines with hard and soft paradigms did not differ in mean size. Class size alsovaried by Carnegie classification. From smallest to largest by classificationwere liberal arts, other institutional types, public two-year, comprehensive, doc-toral, and research institutions. The picture for instructional method was lessclear because several variable levels were equivalent in size. The number oflevels of instructional method was reduced by combining similar class levels.For the remaining four main effects variables, institutional size accounted for6.9%of variance in class size, pure versus applied accounted for 3.4%, lifeversus inanimate for 1.8%, and institutional control for 1.2%.

Instructional Method and Class Size

All other things being equal, the type of classroom interaction that occursbetween teachers and students should logically limit class size. In fact, instruc-tional method should probably be the most important variable in determiningclass size and should exceed disciplinary content, type and size of institution,student level, and all other relevant descriptive information in creating logical,pedagogical ceilings. Based on this assumption, the first data reductive tech-nique employed was one to determine where and to what extent different teach-ing methods were associated with class-size differences. Faculty responding tothe interview form identified the primary instructional method employed ineach class among nine alternatives: lecture; seminar; discussion group or classpresentations; lab, clinic, or problem solving; apprenticeship, internship, fieldwork, or field trips; role-playing, simulation, or other performance (e.g., art,music, drama); TV or radio; group projects; or cooperative learning groups.

Contrary to expectation, instructional method was the second most informa-

624 CHATMAN

tive descriptive variable in explaining class-size differences. The most impor-tant variable was institutional size, which accounted for 6.9% of variance inclass size. Instructional method accounted for 6.3% of variance. Using an ex-tremely conservative Tukey's standardized range test with an alpha level of.0001 to compare size by instructional methods and find types that did notdiffer significantly, three clusters of similar sizes were identified (see Table 3).The first cluster of classes that was not significantly different was composed ofseminars, labs, clinical activities, apprenticeships, internships, and other instruc-tional methods relying on student activity. The second cluster was of groupprojects, cooperative learning groups, discussion or class presentation, and TVor radio courses where the instructional interchange is more likely shared byfaculty and students. The third cluster was of only one type of instruction,lecture, where the faculty clearly direct the interchange. A second main effectsanalysis by these three clusters found none to be equivalent in size. Actually,the three clusters were not perfectly mutually exclusive. The TV or radiocourses and group projects were few in number with large variance and werenot significantly different from any of the other methods. TV and radio coursesand group projects were placed in the second cluster based on mean size. Thisreduction in variable class levels from nine to three was associated without lossof explanatory power (both 6.3%).

Multiple Analysis of Variance

Based on the results of data reduction and simplification from the main ef-fects models, the resulting six variables were asserted as main and interactiveeffects in a general linear model analysis of class size (natural logarithm). Theresults are presented in Table 4. Table 4 displays variance explained by thevariable independently (as a single main effect), cumulatively and incrementally(Type I SS), and uniquely (Type III SS). Also shown in Table 4 are the associ-ated F values, degrees of freedom, and probability levels. Of course, the verylarge degrees of freedom produce highly significant differences of almost allcomparisons. If variance explained independently and incrementally is used in-stead to identify the most important factors, the analysis might stop where theline has been added to the list, after Carnegie classification and before institu-tional control, which means that instructional method, the Biglan dimensions ofpure versus applied and life versus inanimate, institutional size, and Carnegieclassification help to explain class size differences. It should also be noted thatinteractive terms were collectively unimportant as a complete factorial modelaccounted for 3.9% more variance than the simple multiple model.

The order of entry respected pedagogical issues before parochial issues. If itis recognized that the size of a class is a variable under the control of theinstitution, then differences in size beyond those dependent on the content of

LOWER-DIVISION CLASS SIZE 625

TABLE 3. Variance Explained by Main Effects Measures

Percentile

Instructional Method1. Apprenticeship,Internships, Field Work, orField Trips2. Role Play, Simulations,or Other PerformanceActivities3. Seminar4. Lab, Clinical, orProblem Solving

5. Group Projects6. Cooperative LearningGroups7. Discussion Groups orClass Presentation8. TV or Radio Courses

9. LectureInstructional MethodCollapsed

1. Lab, Performance andSeminar (1, 2, 3, 4 above)2. Discussion (5, 6, 7, 8above)3. Lecture

Institutional Size1. 1-6312.632-1,4853. 1,486-3,5644. 3,565-7,7885. Above 7,788

Carnegie Classification1. Liberal Arts2. Other3. Public Two-year4. Comprehensive5. Doctoral6. Research

Institutional Control1. Private2. Public

Pure versus Applied — BiglanDimension

1. Applied2. Pure

Mean

2.84

2.892.92

2.92

3.04

3.07

3.083.15

3.29

2.91

3.083.29

2.802.993.123.273.40

2.993.083.143.303.403.55

3.053.22

3.063.28

SD

0.79

0.650.61

0.58

0.50

0.45

0.490.70

0.59

0.61

0.490.59

0.680.600.560.520.63

0.630.630.520.600.730.90

0.660.57

0.620.56

N R2

6.3%

163

1,155903

3,256

108

758

3,60778

19,740

6.3%

5,477

4,55119,740

6.9%1,3884,0539,2189,3635,746

4.2%2,1932,110

16,7186,0031,5901,154

1.2%5,871

23,897

3.4%12,66717,101

F

251

1,001

551

261

366

1.045

df8

2

4

5

1

1

Pr > F NSD*

0.0001

8, 6,5,4,3,2,1

8, 5,4,3,2,18, 5,4,3,2,1

8, 5,4,3,2,1

9,8,7,6,5,4,3,2,1

8,7,6,5, 1

8,7,6,59,8,7,6,5,4,3,2,1

9,8, 5

0.0001

1

23

0.00011

23

45

0.00011

23

45

60.0001

12

0.00011

2

626 CHATMAN

TABLE 3. (Continued)

Life versus Inanimate —Biglan Dimension

1. Inanimate2. Life

Hard versus Soft Paradigm —Biglan Dimension

1. Hard2. Soft

PercentileMean

3.143.32

3.203.18

SD

0.580.62

0.630.58

N

21,6358,133

10,04919,719

R2

1.8%

0.0%

F

534

5

df

\

1

Pr > F

0.0001

2

0.0254

2

NSD*

1

I

*Not significantly different using Tukey's studentized range test with alpha = .0001. Also note thatclass level labels are repeated.

the course and the instructional method selected to deliver that content aredifferences under the control of the institution. In other words, institution sizewas a significant variable after method of instruction and Biglan dimensions,which means that smaller institutions have decided to offer smaller classes.Similarly, the fact that liberal arts and two-year public institutions have smallerclasses means that they have decided to offer smaller classes. Residual varianceat the lower-division level after course content and instructional method is par-

TABLE 4. Variance Explained by Multiple Model Main Effects Measures

Variance Explained

Independent Cumulative Incremental Unique(%) (%) (%) (%) F df Pr > F

InstructionalMethod CollapsedPure versusApplied — BiglanDimensionLife versusInanimate — BiglanDimensionInstitutional SizeCarnegieClassification

InstitutionalControl— Publicversus Private

6.3

3.9

1.86.9

4.2

1.2

6.3

8.4

10.316.5

17.9

18.4

6.3

2.1

1.96.3

1.4

0.4

4.5

1.9

1.92.7

1.8

0.4

1,148

764

669570

103

150

2

1

14

5

1

0.0001

0.0001

0.00010.0001

0.0001

0.0001

Note: A complete factorial model, including all main effects and interactions, accounted for 22.3%of variance (3.9% more).

LOWER-DIVISION CLASS SIZE 627

tially explained by institution size and Carnegie classification—two variablesrelated to size with no pedagogical basis.

Table 5 displays descriptive statistics and percentile distribution for class sizeby instructional method and the Biglan dimensions of pure or applied and lifeor inanimate studies. Sorted by mean size the general multiple variable patternssupport the main effects results, but extend the trends to a point where thelargest mean classes are twice as large as the smallest. The largest classes arelectures in theoretical life systems disciplines such as botany, physiology, psy-chology, and sociology. The smallest classes are labs or seminars in appliedinanimate fields such as engineering, computer science, and economics. Be-tween these extremes is a pattern generally reflecting the pattern of life systems

TABLE 5. Descriptive Statistics for Classes by Instructional Method andDisciplinary Characteristics

Biglan Dimension

Pure orInstructional Method Applied

Lecture

Discussion

Lab, Performance

Applied

Pure

Applied

Pure

, Seminar Applied

Pure

Life orInanimate

LifeInanimateLifeInanimateLifeInanimateLifeInanimateLifeInanimateLifeInanimate

PercentileMean

30.728.441.931.225.223.530.423.823.219.827.623.7

10% 25%1211201512101413108

1210

201625211715201913121815

50%

252535272420282420182422

75%

353245353026372727243226

90%

504570453835503440304535

N

1,9345,6363,5688,602

762799330

2,6601,0622,474

4771,464

Sorted by Mean SizeLectureLectureLectureDiscussionLectureLab, PerformanceDiscussionDiscussionLab, PerformanceDiscussionLab, PerformanceLab, Performance

PurePureAppliedPureApplied

, Seminar PureAppliedPure

, Seminar PureApplied

, Seminar Applied, Seminar Applied

LifeInanimateLifeLifeInanimateLifeLifeInanimateInanimateInanimateLifeInanimate

41.931.230.730.428.427.625.223.823.723.523.219.8

20151214111212131010108

252120201618171915151312

352725282524242422202018

453535373232302726262724

704550504545383435354030

3,5688,6021,934

3305,636

477762

2,6601,464

7991,0622,474

Note: Classes with enrollment of less than 2 were eliminated.Source: 1992-93 National Survey of Postsecondary Faculty

628 CHATMAN

being larger than inanimate systems, theoretical fields being larger than appliedfields, and lectures being larger than discussions that are larger than seminars.

Graduate Teaching Assistants

The exclusion of teaching assistants could have been associated with a sys-tematic exclusion of classes taught at doctoral and research institutions. Espe-cially at the lower-division level, many classes at large research institutions aretaught by graduate teaching assistants. Based on study of class sizes at a largepublic research institution in the Midwest, a majority of lower-division lecture,recitation, seminar, discussion, and laboratory classes are taught by graduateteaching assistants. Furthermore, the classes taught by graduate teaching assis-tants tend to be smaller (Mullen, 1994). For the NSOPF survey data overall,about 26% of lower-division classes at research universities and 18% of lower-division classes at doctoral universities were reported to have teaching assis-tants. Teaching assistants were also common at liberal arts institutions (11%of lower-division classes), comprehensive institutions (7% of lower-divisionclasses), and public two-year institutions (4% of lower-division classes). Fur-thermore, this pattern might suggest different definitions of teaching assistantsat different institutional levels. Considered collectively, these results suggestthat comparisons of class size by institutional type will probably inflate meanclass size at research institutions. The extent of this inflation is unknown.

CONCLUSIONS

This article began with the assertion that there are very few comparativestandards for postsecondary education that recognize even the most simplisticdistinctions of academic discipline and course level or type of course. Further-more, it was asserted that this lack of information severely limits the ability ofadministrators to defend current practice or identify a direction for change. Inthe absence of standards, the 1990s is as described by Zemsky and Massy(1995), a time of making institutions less costly by giving them less money.Perhaps the results presented here can help to establish reasonable performancestandards.

Class size was determined to be a function of both institutional and coursecharacteristics with institutional characteristics dominating. Classes that de-manded active student participation were smaller as were classes at smallerinstitutions, liberal arts and two-year institutions. Classes were also smaller inapplied fields and fields that focused on living organisms. Other factors wererelatively unimportant. Notably, public or private institutional control andwhether discipline areas assert clear central paradigms (hard) explained little ofthe differences in class size observed. Likewise, more complicated models in-

LOWER-DIVISION CLASS SIZE 629

volving interactions were unnecessary. Based on these results, a series of class-size distributions was reported to help administrators evaluate local course of-ferings. The results also raised the question of why mean lower-division classsize varied from smallest at liberal arts institutions to largest at research institu-tions.

There are problems with this study. First, it is clear that there was tremen-dous overlap in all class-size distributions even when important variables werecontrolled. This may have been the result of much subjective judgment in as-signing the instructional method label and that discretion likely contributed tooverlap and error variance. There were probably also cases of misapplication ofsurvey definitions for courses with mixed types and shared responsibility. How-ever, there is exceedingly little known about the interplay of instructionalmethod, course content, and class size, so the tremendous overlap may be valid.A second problem with this study is the exclusion of graduate teaching assis-tants. Because they were excluded, institutional size and Carnegie class com-parisons may suffer from inflated reported class sizes at large research univer-sities and perhaps doctoral institutions. In any event, the sizes reported werethose for classes taught by a broadly described faculty and there were relativelyfew faculty from these institutions excluded. Third, course and institutionalcharacteristics were able to explain only about 18% of variance in class size.Much remains to be explained.

In spite of these limitations, knowledge of comparable figures for class sizeshould be generally useful. A comparable standard provides a frame of refer-ence against which an institution with relatively smaller classes might promotethe fact to its competitive advantage or one with larger classes might argue forreduced loads. Even institutions near the average can argue for maintainingposition in times of limited resources or improving position in better times. Atthe level of an institutional administrator, comparative standards provide anexternal measure with which to evaluate the instructional production claims ofdisciplinary administrators that are not parochial or anecdotal. Institutionalplanners can use comparative standards to predict the resource requirements ofnew programs. Academic researchers might attempt to determine why disciplin-ary differences exist for a given instructional method or follow national trendsin class size to assess response to resource pressures and challenges. These andother applications are possible when there is a valid comparative standard.

REFERENCES

Biglan, A. (1973a). The characteristics of subject matter in different academic areas.Journal of Applied Psychology 57: 195-203.

Biglan, A. (1973b). Relationship between subject matter characteristics and the structureand output of university departments. Journal of Applied Psychology 57: 204-213.

630 CHATMAN

Bloom, A. M. (1983). Differential instructional productivity indices. Research in HigherEducation 18(2): 179-193.

Chatman, S. P. (1995). Comparable standards for credit hour production. Paper presentedat the 35th Annual Forum of the Association for Institutional Research, Boston.

Creswell, J. W., and Bean, J. P. (1981). Research output, socialization, and the Biglanmodel. Research in Higher Education 15(1): 69-91.

Creswell, J. W., and Roskens, R. W. (1981). The Biglan studies of differences amongacademic areas. Review of Higher Education 4(3): 1-16.

Creswell, J. W., Seagren, A. T., and Henry, T. C. (1979). Professional development train-ing needs of department chairpersons: A test of the Biglan model. Planning andChanging 10: 224-237.

Eison, C. L. (1976). The measurement of satisfaction in departmental association atWestern Kentucky University testing the Holland and Biglan models. Unpublisheddoctoral dissertation, University of Kentucky.

Glass, G. V., and Smith, M. L. (1979). Meta-analysis of research on class size andachievement. Educational Evaluation and Policy Analysis 1: 2-16.

Glass, G., Cahen, L., Smith, M. L., and Filby, N. (1982). School Class Size. BeverlyHills, CA: Sage.

Levin, H. M. (1988). Cost-effectiveness and educational policy. Educational Evaluationand Policy Analysis 10: 51-69.

Malaney, G. D. (1986). Differentiation in graduate education. Research in Higher Educa-tion 25(1): 82-96.

Maxwell, N. L., and Lopus, J. S. (1995). A cost effectiveness analysis of large and smallclasses in the university. Educational Evaluation and Policy Analysis 17(2): 167-178.

McKeachie, W. J. (1980). Class size, large classes and multiple sections. Academe 66:24-27.

Muffo, J. A., and Langston, I. W. (1981). Biglan dimensions: Are the perceptions empiri-cally based? Research in Higher Education 15(2): 141-159.

Mullen, R. (1994). On-Campus Teaching Load Responsibility Fall, 1994, 12-28. Officeof Planning and Budget, University of Missouri System, Columbia.

Siegfried, J. J., and Pels, R. (1979). Teaching college economics: A survey. Journal ofEconomic Literature 17(3): 923-969.

Slavin, R. (1990). Class size and student achievement: Is smaller better? ContemporaryEducation 62(1): 6-12.

Smart, J. C., and Elton, C. F. (1975). Goal orientations of academic departments: A testof Biglan's model. Journal of Applied Psychology 60: 580-588.

Smart, J. C., and Ethington, C. A. (1995). Disciplinary and institutional differences inundergraduate education goals. Paper presented at the annual meeting of the Associa-tion for the Study of Higher Education, Orlando.

Smart, J. C., and McLaughlin, G. W. (1978). Reward structures of academic disciplines.Research in Higher Education 8(1): 39-55.

Williams, D. D., Cook, P. F., Quinn, B., and Jensen, R. P. (1985). University class size: Issmaller better? Research in Higher Education 23(3): 307-317.

Zemsky, R., and Massy, W. F. (1995). Toward an understanding of our current predica-ments. Change 27(6): 40-49.

Received May 28, 1996.