Does Academic Research Help or Hurt MBA Programs

8/3/2019 Does Academic Research Help or Hurt MBA Programs

1/20


2/20

32 / Journal of Marketing, September 2008

the other measure, after the effects of the other factors areaccounted for.

Overall, these studies and viewpoints argue that acade-mic research does not improve a schools educational per-formance or its business press ranking. Thus, academicresearch is forced to compete with other business schoolactivities rather than being viewed as the foundation of itsbroad educational mission. When faced with such compet-ing activities, business school administrators will allocatefewer resources to research if they believe that research

serves faculty members interests without benefiting otherconstituents, such as students and recruiters.

Decisions about the role of research in the broader edu-cational mission of business schools are extremely impor-tant to students, faculty members, firms, and society atlarge. Thus, we believe that much more research is neededon this topic. In particular, four key limitations of existingresearch should be addressed. First, extant research focuseson the contemporaneous effects of academic research, pri-marily through cross-sectional, correlational analyses.However, academic research may have long-term payoffs,so it is necessary to evaluate both short-term and long-termeffects. To the best of our knowledge, no existing study

examines the long-term effects of academic research. Sec-ond, no current study considers the multiple constituentsthat business schools serve, including academics from otherbusiness schools, recruiters, and applicants (i.e., potentialstudents); it is not known whether research affects theseconstituents and, if so, how large these effects are. Asadministrators decide how much of their resources to allo-cate to academic research, it would be helpful to knoweffect sizes across these constituents. Third, it is not knownwhether the effects of academic research will vary with thereputation of a business schools parent university. Fourth,

current studies focus on the relationship between teachingand research at the individual level. However, these studiesdo not consider broader, more objective measures of educa-tion performance that may better reflect the overall impactof academic research in business schools. In this study, webegin to address these important limitations and providesome additional evidence on the long-term relationshipbetween academic research and MBA program perfor-mance. We examine MBA program performance in terms ofboth the perceptions of different constituents and the educa-

tional value created. Without such evidence, critics of MBAprograms will continue to assume that their polemicsagainst research are justified.

In the next section, we present a simple framework forevaluating the short-term and long-term effects of academicresearch on the perceptions of important constituents andalso on the educational value for students. Then, wedescribe our modeling approach and the context of ourstudy. Following that, we present a broad set of results onacademic research in business schools and conclude with adiscussion of our findings, their implications, and relatedopportunities for further research.

A Long-Term, Multiple-ConstituentFramework for MBA Programs

MBA programs serve multiple constituents, whose percep-tions are used to evaluate the performance of these pro-grams. In Figure 1, we present a simple framework for thelong-term dynamic relationships among business schoolsresources, activities, performance, and constituents percep-tions of performance. Similar to other organizations, busi-ness schools have a variety of resourcesboth tangible(e.g., physical, financial) and intangible (e.g., brand, intel-

FIGURE 1A Dynamic Model of Educational Institution Performance

ResourcesBrand

FinancialPhysical

ActivitiesResearch

EducationCertification

Researchperformance

Educationperformance

Applicantsperceptions

RecruitersPerceptions

Academicsperceptions

ConstituentsPerceptionsPerformance


3/20

Does Academic Research Help or Hurt MBA Programs? / 33

lectual). These resources support multiple activities in pur-suit of the organizations multiple performance objectives.Actual performance on these different activities sometimescan be measured objectively (e.g., number of faculty publi-cations, students starting salary). However, even whenactivities can be objectively measured, constituents percep-tions are still important. These perceptions feed back tofuture-period resources, resulting in a dynamic frameworkwith both short-term and long-term effects. These effectscan be on constituents perceptions or on objective perfor-

mance measures. Any activity can have either a positive or anegative long-term impact on the performance of anotheractivity. For example, if the American Marketing Associa-tion were to increase its emphasis on training sessions, theobjectives of academics might be marginalized, at least inthe short run. However, over time, if these training activitiesbrought in resources that were spent on increasing activitiesgeared toward academics, there could be a positive long-term effect. Conversely, if these resources were not spent onacademics activities, the long-term effect of training onacademics activities would remain negative.

In business schools, various constituents with differentobjectives compete for the schools resources. Administra-

tors decisions about allocating resources across these com-peting objectives are exacerbated by the lack of a unifyingprofit objective. In for-profit enterprises, all investments canbe ranked by their return on investment. However, in non-profit organizations, the objectives of various constituentscannot be compared similarly (Gallagher and Weinberg1991; Newman and Wallender 1978). Thus, our frameworkand approach may be relevant for other types of nonprofitorganizations with multiple activities and multiple con-stituents. For example, museums not only want to showcasetheir current collections but also want to invest in acquiringor discovering new items. Libraries not only loan a portionof their collection but also invest in preserving rare manu-

scripts. Public theaters not only want to expose people to

their productions but also want to create new works of art(Voss, Montoya-Weiss, and Voss 2006). Different con-stituents will have different levels of importance attached tothese activities. In Table 1, we illustrate the diversity of con-stituents for nine types of nonprofit organizations anddivide these constituents into internal, external, and mixedconstituents. Internal constituents are those who develop ordeliver the organizations services. External constituents arethose who consume the organizations services. Mixed con-stituents are those who are partially involved in developingor delivering the organizations services and partiallyinvolved in consuming those services. In business schools,internal constituents include professors, instructors, andadministrators; external constituents include recruiters,potential students, and consumers of faculty research (e.g.,businesses, consultants, government agencies); and mixedconstituents include current students, who consume facultyteaching and contribute to the consumption of other stu-dents educational experience. Other mixed constituents aretrustees and donors, who also produce and consume ele-ments of a business schools services.

Impact of Academic Research on Multiple

Constituents Perceptions

The impact of academic research on performance is realizedthrough both short-term and long-term effects. A potentialshort-term effect is the impact of academic research on aca-demics perceptions at other business schools. Academicresearch is likely to have a positive impact on academicsperceptions because they decide which research is valuableenough to be published during the review process. Deansand department chairs, who provide the data for academicsperceptions to the business press, are exposed to their ownfaculty members in-depth knowledge of other schoolsresearch activities. Faculty may communicate this knowl-

edge to them through formal reviews and informal conver-

TABLE 1

Nonprofit Organizations and Examples of Their Constituents

Organization Internal Constituents Mixed Constituents External Constituents

Business schools Professors, instructors,administrators

Applicants, alumni, trustees,donors

Recruiters, consumers ofresearch (e.g., consulting

firms)

Charitable/relief organizations Administrators, employees Donors, funding agencies Aid recipients

Churches Clergy Members Attendees

Hospitals Doctors, nurses, administrators Residents, interns Patients

Legal aid societies Lawyers Funding agencies Clients, potential clients

Museums Administrators, employees Donors, funding agencies Visitors

Performing arts organizations Director, conductor,long-term performers

Contract performers,season ticket holders, donors

Single-ticket holders

Political organizations Politicians, administrators Donors Voters, media

Professional associations Administrators, employees Members Media


4/20


sations. Formal reviews sometimes compare the researchperformance of one school with that of other schools, eitherat the department level or at the school level. In addition tofaculty input, deans with research backgrounds and depart-ment chairs may have firsthand knowledge of other schoolsresearch performance, especially within their own disci-plines. Finally, deans also receive promotional materialabout research productivity from other business schools.Because of these sources of information, we expect bothpositive short-term and positive long-term effects of acade-

mic research on academics perceptions of other MBAprograms.

Long-term effects of academic research are likely topredominate with other constituents, such as recruiters andapplicants. Because recruiters and applicants will have lim-ited direct contact with most faculty members research, wedo not expect any short-term effect of academic research onthe perceptions of these constituents. However, graduatestudents exposed to the latest research may be betteremployees, thus motivating recruiters to hire more studentsfrom that school. Better placements will then attract higher-quality applicants to that school in the future. Thus, overtime, we expect that academic research will have a positive

long-term effect on recruiters and applicants perceptions.On the basis of this discussion, we propose severalhypotheses.

H1: There is a positive short-term effect of academic researchon academics perceptions of MBA programs.

H2: There are positive long-term effects of academic researchon academics, recruiters, and applicants perceptions ofMBA programs.

H3: The duration of the long-term effects of academic researchon perceptions is shorter for academics than for recruitersand applicants.

Impact of Academic Research on Education

Performance in Business SchoolsAcademic research focuses on knowledge creation, butanother key objective in business schools is knowledge dis-semination through the education of students. Figure 1 indi-cates two effects of research on education performance: ashort-term effect and a long-term effect. First, a facultymembers research can enhance his or her knowledge of thesubject and his or her ability to motivate students interest inthat subject. Moreover, a researchers ability to teach theprocess of knowledge discovery embodied in researchshould be valuable to his or her students. Thus, the educa-tional value created by a business school should be posi-tively affected by the direct interactions between research-

active faculty and students. However, the impact of researchon education performance may also have negative short-term effects. A faculty members time is a scarce resource.Thus, too much time devoted to research will detract fromthe time he or she can devote to educating students. In addi-tion, the skills required to be an effective researcher may bedifferent from the skills required to be an effective teacher.Thus, focusing on academic research may detract from abusiness schools educational objectives. Based on the largenumber of empirical investigations that find no relationship

between academic research and teaching effectiveness,these positive and negative effects seem to balance eachother out, at least in the short run.

Second, over the long run, academic research shouldpromote a researchers teaching ability because new knowl-edge is typically relevant for several years. Furthermore,this new knowledge diffuses through researchers socialnetworks. Researchers interact closely with other facultymembers, maximizing knowledge diffusion and educationalbenefits. Over time, we also expect that academic research

will have a positive effect on the perceptions of the variousconstituents. This enhanced reputation is likely to generatefinancial and physical resources, further contributing to thelong-term effect. Thus:

H4: There is a positive long-term effect of academic researchon the education performance of MBA programs.

Effect of Prior Brand Reputation

Brand reputation is formed through word of mouth, pastperformance, and personal experiences. Previous researchhas found that customers are more likely to incorporateinformation that is consistent with their beliefs (Boulding,

Kalra, and Staelin 1999; Hoch and Ha 1986; Lord, Lepper,and Ross 1979). Improvements in research will be moreconsistent with peoples beliefs about business schools inhigh-reputation universities. Therefore, we expect an inter-action effect between research and university reputation onconstituents perceptions. According to Figure 1, this willalso result in higher future resources. Higher resources willlikely improve the education-related activities, which inturn will improve the education performance of the programover the long run. Thus:

H5: The long-term effects of an increase in academic researchon the education performance of an MBA program aregreater for programs in higher-reputation universities than

for those in lower-reputation universities.

Apart from a larger effect size for information that is con-sistent with prior reputation, the updating process is alsolikely to be faster (Geers and Lassiter 1999; Wilson et al.1989). Thus:

H6: The duration of the long-term effect of an increase in aca-demic research on the education performance of an MBAprogram is shorter for programs in higher-reputation uni-versities than for those in lower-reputation universities.

Definitions

Short-Term and Long-Term Effects

The hypotheses propose the short-term and long-termeffects of academic research on perceptions of multipleconstituents and on the MBA programs impact on educa-tion performance. The short-term effect of academicresearch is its impact in the current period (i.e., same yearin this study). The long-term effect is its cumulative impactover an infinite time horizon. Carryover duration is the timeneeded to reach a prespecified percentage of the long-term


5/20


1We also examine size of the program (i.e., student enrollment)as another control variable. However, we dropped this variablebecause it is highly correlated with wealth (.89).

effect. Typically, studies report the 90% carryover duration(e.g., Clarke 1976; Mitra and Golder 2006; Van Heerde,Mela, and Manchanda 2004).

Education Performance of Business Schools

A business schools inputs consist of its incoming studentsand the schools resources used to educate those students.An important output of a business school is the graduates ofits programs. The differential between the levels (e.g., qual-ity) of input and output refers to that portion of studentgrowth or development that can be attributed to the specificeducational experience offered by the business school(Lockheed and Hanushek 1994, p. 1779; see also Tracy andWaldfogel 1997). Therefore, we measure the differentialbetween the input and the output as the value-added bybusiness schools and define it as education performance.The education literature describes two related concepts ofeducation performanceeffectiveness and efficiency(Cowan 1985; Hanushek 1986). When inputs are repre-sented in nonmonetary units, the differential is referred toas effectiveness. When inputs are represented in monetaryunits, the differential between input and output reflectsefficiency.

ModelIn line with our conceptual framework described in Figure1, the current performance perceptions of an MBA programare dependent on both current academic research and theprior-period performance perceptions of the MBA program.In particular, the link between prior-period performance andcurrent performance is the result of multiple feedback pro-cesses (represented by dotted links in Figure 1). Thus:

(1) Pijt = b0ij + b1ijPij(t 1) + b2ijRit + ijt,

where MBA program is performance perception (P) of

constituency j at time t is related to academic research (R)at time t and P at time t 1.Equation 1 signifies either a serial correlation or a state-

dependent carryover process. Jacobson (1990) shows thatadding a lagged independent variable in Equation 1 enablesthe appropriate interpretation of its parameters. Specifically,if the significance of b1ij is due to serial correlation, thecoefficient of the lagged independent variable term is of theopposite sign (i.e., compared with the coefficient of theindependent variable). If not (i.e., if it is zero or positive),we can interpret the parameters of the model in terms of acarryover process. Thus, to distinguish between these pro-cesses, we add a lagged academic research variable. Toobtain the true effects of academic research, we controlfor other variables, such as the institutions wealth andtuition.1 As a result, we modify Equation 1 as follows:

(2) Pijt = b0ij + b1ijPij(t 1) + b2ijRit + b3ijRi(t 1) + b4ijCit + ijt,

where Ci denotes the control variables.

Equation 2 is of the form of a partial adjustment model,which has been frequently used to study long-term effects(e.g., Mitra and Golder 2006; Tellis, Chandy, andThaivanich 2000; Van Heerde, Mela, and Manchanda2004). The model parameters can then be used to computethe short-term effect, the long-term effect, and the carryoverduration of academic research. The short-term effect of aca-demic research on performance objective j for MBA pro-gram i is

(2a) Es = b2ij,

and the long-term effect is

(2b) El = (b2ij + b3ij)/(1 b1ij).

We can also test for the significance of E s and El on thebasis of the covariance matrix of the parameters. Further-more, we can compute the duration of the carryover of aca-demic research on performance as follows:

The p% carryover duration denotes the time during whichp% of the cumulative long-term effects are realized (fordetails, see Clarke 1976).

There are other benefits of Equation 2 as well. First, thelagged independent variable term (b3ij) can reveal nonmo-notonic carryover structures when b3ij > b2ij (Clarke 1976).Second, adding the lagged independent variable term hasbeen shown to reduce data interval bias in distributed lagmodels (Russell 1988). Finally, it has been established thatthe behavioral process that leads to a distributed lag struc-ture is a rational response to the uncertainty in the causal

variable (Lutkepohl 1984; McLaren 1979). In our case, thismodel form is credible because many people (e.g.,recruiters, potential students) do not know the causalvariable academic research with certainty.

Empirical Models

We present our approach for empirically evaluating ourhypotheses and estimating our models. Note that H1, H2,and H3 are based on the effects of academic research onperceptions of three types of constituents. H4 is based onthe effects of academic research on education performance,and H5 and H6 are based on how the size and duration ofthis effect are influenced by university reputation. We spec-

ify the models that link academic research with multipleconstituents perceptions and the education performance ofa business school. We then expand the education perfor-mance model to include the hierarchical effects of reputa-tion, as proposed in H5 and H6. Afterward, we discuss esti-mation details.

Multiple constituents perception model. For H1H3, weestimate separate equations for the effect of academicresearch on the perceptions of different constituents. Thus:

(2c) p% carryover duration =

b1i

1

1

+

ln( ) lnp jj 2ij 3ij

2ij 3ij

1ij

b b

b b

b

+

+

ln

.


6/20


(3) Pijt = b0j + b1jPij(t 1) + b2jRit + b3jRi(t 1) + b4jCit + ijt,

ijt = uij + vtj + eijt,

where i refers to business school; j refers to constituenttype; t refers to time; uij and vtj are random variables thatpartition the error in terms of the business school and time,respectively; and eijt is a classical error term with zero meanand homoskedastic covariance structure.

We can state H1H3 as follows:

H1: b

2j> 0, where j refers to academics perceptions.

H2: [(b2j + b3j)/(1 b1j)] > 0 for all j.

H3: Carryover duration for academicsperceptions < carryoverduration for recruiters perceptions < carryover durationfor applicants perceptions.

Education performance model. For H5 and H6, we firstestimate a model that links academic research and educa-tion performance of business schools using a homogeneousslope parameter for all business schools. Thus:

(4) Eit = b0 + b1Ei(t 1) + b2Rit + b3Ri(t 1) + b4Cit + it,

it = ui + vt + eit,

where MBA program is education performance (E) at time

t is related to academic research (R) at time t and E at timet 1; ui and vt are random variables that partition the errorin terms of the business school and time, respectively; andeit is a classical error term with zero mean and homoskedas-tic covariance structure.

Using Equation 4, we can state H4 as follows:

H4: [(b2 + b3)/(1 b1)] > 0.

Note that H5 and H6 describe how the short-term andlong-term parameters in Equation 4 are affected by an inter-action between improvement in research over time andreputation of the university to which the business schoolbelongs. Thus, we introduce time-level and cross-section-

level heterogeneity in the parameters of Equation 4 througha two-level hierarchical model that is simultaneously esti-mated. This approach has been usefully applied in manyprior studies (e.g., Anderson and Salisbury 2003; Mitra andGolder 2006; Steenkamp, Ter Hofstede, and Wedel 1999).

The lowest hierarchy in the data is that of businessschool year. Therefore, the first-level model is estimated forevery business school year unit. Thus:

(5a) Eit = b0 + (b1 + 1iImprovementt 1)Ei(t 1)

+ (b2 + 2iImprovementt)Rit

+ (b3 + 3iImprovementt 1)Ri(t 1) + b4ijCit + it,

it = ui + vt + eit,where Improvementt = 1 when research increases in time tcompared with (t 1).

For the second-level model, we estimate the parametersof the first-level model that vary over business schools as afunction of the universitys reputation:

(5b) 1i = 10 + 11Reputationi + r1i, where r1i ~ N(0, 12);

2i = 20 + 21Reputationi + r1i, where r2i ~ N(0, 22); and

3i = 30 + 31Reputationi + r1i, where r3i ~ N(0, 32),

2For deriving the variance of the functions that test our hypo-theses, based on the parameters in Equations 3, 4, and 5, we useCramers theorem: If = g(), then Var () = g g, where g = rowvector of the first partial derivatives of function g, g = columnvector of the first partial derivatives of function g, and =variancecovariance matrix of the parameters.

where Reputationi = 1 when business school i is in a high-reputation university.

On the basis of Equations 5a and 5b, we can state H5and H6, respectively, as follows:

and this condition simplifies to

(6) (1 b1 10)(21 + 31) + 11[(b2 + b3) + (20 + 30)] > 0.

H6: Carryover duration for an increase in research for high-reputation schools > carryover duration for an increase inresearch for other schools.

Estimation of Models

We estimate Equations 3 and 4 with a time-series cross-sectional regression method because we have data on manybusiness schools over time. We use the TSCSREG proce-dure of SAS that employs a feasible generalized least

squares method and a two-way random-effects model thatseparates the error into business schoolspecific and time-specific components. There is ample evidence that feasiblegeneralized least squares estimators are asymptoticallyequivalent to generalized least squares estimators and thusare consistent and efficient (Amemiya 1985; Wooldridge2002).

We chose a random-effects model for two reasons. First,because our model contains a lagged dependent variable, afixed-effects model would generate biased estimates of thecoefficients (Nickell 1981). Second, the fixed-effects modelassumes that all business schools are independent. How-ever, schools may share geographical locations, public or

private status, and even some of the same faculty membersor deans during the period of our data. Conversely, arandom-effects model is appropriate for panel data of manybusiness schools that are drawn from a large population(Greene 1990, p. 485).

We estimate Equations 5a and 5b with the HLM 5 pro-gram developed by Raudenbush, Bryk and Congdon (seeBryk and Raudenbush 1992). This program applies a maxi-mum likelihood procedure to generate parameter estimatesthrough an iterative EM (expectationmaximization)algorithm.

Because our hypotheses are based on a function of theestimated parameters, we use Cramers theorem to derive

the variance of these functions.2

This approach enables usto evaluate these hypotheses with significance tests. How-ever, as in prior studies, we cannot statistically test for the

Hb b

b52 20 21 3 30 31

1 10 111:

( ) ( )

( )

+ + + + + + +

+ + +

+

>

( ) ( )

( ),

b b

b2 20 3 30

1 1010


7/20


3Note that USN also reports an overall score based on thesemeasures. However, we do not use this score, because it is contro-versial and there are disagreements on the relative importance ofthe different component measures. Nevertheless, we find signifi-cant short-term and long-term effects of academic research onUSNs overall performance construct.

significance of the duration of the long-term effect (e.g.,Mela, Gupta, and Lehmann 1997; Van Heerde, Mela, andManchanda 2004).

DataTo evaluate our hypotheses, we need longitudinal data formany MBA programs over a long period (Golder 2000).After collecting data from multiple sources and mergingthem into a common data set, we believe that our data are

sufficient for providing both insights into the short-term andlong-term effects of academic research on several MBAprogram objectives and certain asymmetries in theseeffects. Next, we describe the data collection and operation-alization of variables.

Sources and Data Collection

Our initial goal was to collect relevant data for as manyMBA programs as possible. Our primary data are collectedfrom U.S. News & World Report (hereinafter, USN). TheUSN data have been reported every year since 1990, arewell known, and have been used in previous organizationaltheory research (DAveni 1996; Trieschmann et al. 2000).

Each year, USNsurveys (1) deans or program directors anddepartment chairs or faculty members and (2) a nationwiderandom sample of recruiters to determine their perceptionsof MBA programs. The survey asks these two groups ofrespondents to rate the quality of each program on a Likertscale from marginal (1) to outstanding (5). If a respon-dent is unfamiliar with any program, he or she has theoption of indicating dont know. The results of this surveyfor the top-50 programs are reported in the following year.For example, in April 1990, USNpublished the informationon MBA programs based on the 1989 survey results. Giventhe reputation ofUSN, survey response rates are reported tobe as high as 60%70%.

In addition to academics and recruiters perceptions,USN reports other measures, including average GraduateManagement Admission Test (GMAT) scores and under-graduate grade point averages of incoming students, accep-tance rate of each business school, average starting salaryafter graduation, percentage of graduates employed threemonths after graduation, out-of-state tuition and fees, andfull-time enrollment.3 Subsequently, we explain how we usethese measures to operationalize our variables.

For academic research, we use University of Texas atDallas data on research productivity (see http://citm.utdallas.edu/utdrankings). University of Texas at Dallastracks research contributions based on publications in 24leading journals across major business disciplines. The

database contains author and institutional affiliations for allarticles published in these journals since 1990. Author affil-iations are recorded at the time of publication. Our primary

4We use other cutoffs (e.g., top 15, top 25) as alternative mea-sures and find that our results are robust to these changes.

5We also use business school endowment as an alternative mea-sure of wealth. We are unable to obtain this measure for 11 of 57schools, and it is strongly and positively correlated with universityendowment (.843) for the remaining schools.

measure of academic research is the number of articlesweighted by the number of authors of an article. For exam-ple, a single-author article results in the affiliated schoolbeing credited with a score of 1. If there are multipleauthors, the school gets a score of 1/n for each facultymember, where n is the number of authors. Because theUniversity of Texas at Dallas data do not include 1989 (thefirst year of the data), we manually collect this informationusing the same set of journals and methodology.

For reputation of a university, we collect the overall

ranking of national universities in USN. Note that theserankings are collected separately from different people thanthose providing the business school rankings. Specifically,we compute the average rank of a university during the19892006 period. We consider the top-20 universitieshigher-reputation universities.4 For wealth, we use data onuniversity endowment from the annual survey of theNational Association of College and University BusinessOfficers. This survey has been conducted every year since2001 (for details, see http://www.nacubo.org/x2376.xml).On the basis of these surveys, we use the average endow-ment of the universities during the 20012006 period torepresent the wealth of a university.5

Operationalizing Performance Perceptions

The dependent variables in Equation 3 are the performanceperceptions of academics, recruiters, and applicants. Acade-mics and recruiters evaluations can be obtained directlyfrom the USNdata. However, through 2000, USNreportedperformance perceptions as ranks. Since 2001, performanceperceptions of academics and recruiters are reported on ascale of 1 to 4. To achieve uniformity, we convert the per-ception scores since 2001 to their respective ranks. Usingparametric tests on rank-transformed data provides robustresults and performs as well as cardinal data, particularlyfor relatively small sample sizes (Conover and Iman 1981;

Hettmansperger 1979; Iman and Conover 1979). They havealso been widely used by researchers in business and otherdisciplines (Banker, Davis, and Slaughter 1998; Epley1997; Walther 1997). However, a drawback of using ranksis the interpretation of the effect size in terms of ordinalnumbers instead of units of measurement.

Unlike measures of academics and recruiters percep-tions, there are no direct measures of the performance per-ceptions of business schools applicants. However, theacceptance rate of a business school can be considered anobjective measure of applicants perceptions toward aschool. The acceptance rate conveys the applicantsdemandfor a business school relative to a fairly stable capacity. As a

result, acceptance rates are likely to decrease as applicantsperceptions improve and the number of applicationsincreases.


8/20


6Subsequently, we use an alternative measure of student quality(acceptance rate) to demonstrate the robustness of our results.

7Subsequently, we use an alternative measure of education per-formance (students instructor ratings in two schools) to demon-strate the robustness of our results.

Operationalizing Education Performance

The dependent variables in Equations 4 and 5 are related tothe education performance of a business school in terms ofvalue added to its students. In the Definitions section, wereferred to the two related dimensions of education perfor-manceeffectiveness and efficiency. In line with our defi-nitions, we employ three measures of education perfor-mance. We operationalize effectiveness as the differentialbetween output and nonmonetary input, efficiency as thedifferential between output and monetary input, and overalleducation performance as the differential between outputand both monetary and nonmonetary inputs.

We measure the output of a business school in terms ofthe expected salary obtained by a graduating class. Weobtain the expected salary by multiplying the average start-ing salary by the percentage of graduates employed withinthree months of graduation. We convert the average startingsalary for each year into 1998 dollars (i.e., the median year)to account for inflation. We use the gross domestic productdeflator reported by the National Aeronautics and SpaceAdministration (http://cost.jsc.nasa.gov/inflateGDP.html).Next, we consider two types of inputsnonmonetary andmonetary.

For measuring the effectiveness dimension of educationperformance of business schools, we consider the primarynonmonetary input (i.e., the quality of the students beforereceiving the business school education). We use GMATscores as a measure of average student quality.6 Because thetypical time required for an MBA program is two years, weobtain the relevant GMAT score of each graduating class bylagging the reported GMAT score by two years. For mea-suring the efficiency dimension of education performance,we consider the primary monetary inputs for businessschools (i.e., the tuition and fees students paid, as well asother finances from nonstudent sources, such as alumnidonations and public grants). Tuition and fees are reported

by USN. However, donations and grants at the businessschool level are not publicly reported. Therefore, we use theendowment of the university to control for nontuition mone-tary input. We convert the tuition for each year and averageendowment into 1998 dollars. Finally, for overall educationperformance of a business school, we consider both themonetary and the nonmonetary inputs (i.e., tuition, endow-ment, and GMAT score).

Our specific measures of education performance areinspired by a rich tradition in economics, accounting, andfinance of using residual-based measures of performance(e.g., Bartelsman and Dhrymes 1998; Griliches 1996;Schoar 2002).7 For example, in economics, the well-known

Solow residual is a number describing empirical productiv-ity growth in an economy from year to year (Arrow et al.1961; Hall 1988). Similarly, in finance, accounting, andmarketing, event studies measure the impact of events using

8Because we do not find any significant time trend afteraccounting for the year-level fixed effects, we do not include atrend variable in these models.

residuals of stock returnsthat is, the part of stock returnsnot explained by the market index return (e.g., Brown andWarner 1985; Fama and French 1992; Markovitch andGolder 2008). In a similar spirit, we regress the expectedsalary of the graduating class (output) with the appropriateinputs and obtain the residuals of these regressions. Theseresiduals (i.e., the differentials between the actual expectedsalary and the fitted expected salary, based on these regres-sions) provide our three measures of education perfor-manceeffectiveness, efficiency, and overall performance.8

The residual in any regression captures a combinationof random error and any omitted factors. If a businessschools output is more than what is expected on the basisof its input, the residual is positive. This outcome could bedue to either random error or omitted factors that includethe educational experience provided by the business school.

Our measures of education performance have threeimportant characteristics. First, although we use the residu-als as measures of business schools education performance,readers may prefer to interpret the results in terms of stu-dents excess salary (i.e., the portion of salary not explainedby monetary inputs, nonmonetary inputs, or both). Becausewe are primarily interested in understanding the effects of

academic research, students excess salary is an importantoutcome in itself to evaluate. Second, there are temporaldifferences between the input and the educational experi-ence as well as the output. For example, students take theGMAT before beginning their educational experience, andplacement occurs after both the GMAT and some or all oftheir educational experience. Third, if there are underlyingvariables driving both the input and the educational experi-ence, the residual nature of our measures makes ourapproach conservative. In these cases, the residual is lesslikely to contain information on an underlying variable thatis regressed out as part of the input.

Our three measures of education performance (i.e.,

effectiveness, efficiency, and overall performance) focus oninputs and outputs of the students educational experience.An important component of the education experience isclassroom instruction. Therefore, in the validation section,we report the correlation between our education perfor-mance measures and instructor ratings from students in twomajor U.S. business schools. In addition, we show the long-term impact of academic research on these students ratings.

Our overall data set contains annual observations of theweighted and raw number of articles in 24 peer-reviewed journals, performance perceptions of academics andrecruiters, applicantsacceptance rates, GMAT scores, start-ing salary, employment status of graduates (three monthsafter graduation), and tuition for 57 MBA programs over a

period of 18 years (19892006). On the basis of these data,we compute three measures of education performance(effectiveness, efficiency, and overall performance) forthese programs. In addition, we have the average endow-ment and a categorical measure of reputation for each busi-ness schools parent university. Because we have variables


9/20


9Not all time series are 17 years, because some schools are notin the top-50 MBA programs every year.

lagged by one year, for our analysis, we use a time series upto 17 years for these 57 business schools.9

ResultsWe begin by presenting descriptive results of our data.Then, we discuss the model results and our findings on eachhypothesis. Finally, we examine the robustness of ourresults to certain modeling and data assumptions usingalternative model specifications discussed in the model sec-

tion. In this section, we also use additional data to validateour measures of education performance in business schools.

Descriptive Results

Prior research has examined the correlations of researchwith both teaching and MBA program performance.Although correlations cannot be used to draw inferencesabout causal effects, in Table 2, we present correlationsbetween academic research and performance perceptions ofacademics, recruiters, and applicants. Likewise, in Table 3,we present correlations between academic research and ourthree measures of education performance.

Across all MBA programs over the 19892006 period,

the average correlations between academic research andperformance perceptions of academics, recruiters, andapplicants are .64, .51, and .39, respectively. Negativecorrelations mean that more research is associated withranks closer to one and smaller acceptance rates. Previousresearch has not provided any benchmarks for comparison,

10However, through an empirically derived structural model ofselected organization-level variables, Trieschman and colleagues(2000) find no significant relationship between research and over-all MBA program performance in the presence of other variables.

though it is not surprising that the highest correlation isbetween academic research and academics performanceperceptions. However, these findings are similar to studiesthat report correlations between academic research andpopular press measures of overall MBA program rankings(Siemens et al. 2005; Trieschmann et al. 2000).10 Notably,we find that the correlations between academic research andapplicants acceptance rate are becoming more negativeover time, but there is no time trend in the correlations withother constituents perceptions.

The average correlations of academic research with ourthree education performance measureseffectiveness, effi-ciency, and overall performanceare .53, .34, and .34,respectively. These results contrast with existing literature,which finds academic research and teaching performance tobe essentially unrelated (Feldman 1987, p. 275; see alsoHattie and Marsh 1996). A potential explanation for the dif-ferent results is that previous studies focused on short-termeffects within individual classrooms, whereas we focus oneducation performance of the entire MBA program.Because research may have long-term effects on education,this broader measure and longer time frame may be betterfor capturing the overall impact of a business schools

research culture.Before estimating our models, and because we hypothe-

size that academic research drives changes in performance,we first must establish that academic research itself is

TABLE 2

Correlations of Academic Research and Constituents Performance Perceptions

Academics Recruiters ApplicantsYear Perceptions Rank a Perceptions Ranka Acceptance Rateb

2006 .59 .47 .482005 .60 .40 .502004 .74 .64 .602003 .70 .57 .522002 .67 .47 .532001 .64 .51 .522000 .69 .61 .571999 .71 .63 .601998 .69 .56 .611997 .65 .59 .401996 .68 .56 .231995 .61 .54 .271994 .69 .46 .31

1993 .59 .50 .061992 .64 .45 .361991 .56 .37 .221990 .54 .43 .101989 .59 .38 .19Average .64 .51 .39Linear trendc Not significant Not significant Negative (p< .01)

aMeasured in ordinal rank; a negative correlation indicates that more academic research is associated with a better rank.bMeasured as a percentage of the number of applications received; a negative correlation indicates that more academic research is associatedwith a lower acceptance rate.

cNegative trend denotes a higher negative correlation over time; more academic research is more highly associated with better academics andrecruiters perception rank and lower acceptance rate over time.


10/20


TABLE 3

Correlations of Academic Research and Education Performance

Educational Educational OverallYear Effectivenessa Efficiencyb Performancec

2006 .46 .39 .342005 .48 .34 .292004 .56 .42 .372003 .46 .31 .242002 .47 .29 .25

2001 .47 .31 .302000 .57 .47 .431999 .68 .58 .601998 .57 .47 .461997 .64 .35 .431996 .60 .46 .511995 .59 .41 .481994 .65 .27 .351993 .44 .17 .221992 .37 .13 .021991 .51 .21 .101990 N.A.d .26 N.A.1989 N.A.d .19 N.A.Average .53 .34 .34Linear trende Not significant Not significant Not significant

aMeasured as a students starting annual salary in excess of that predicted by the students GMAT score, in dollars.bMeasured as a students starting annual salary in excess of that predicted by tuition and university endowment, in dollars.cMeasured as a students starting annual salary in excess of that predicted by the students GMAT score and tuition, as well as universityendowment, in dollars.

dWe cannot compute the effectiveness and overall performance for 1989 and 1990, because relevant GMAT figures are not available.eFor the absolute value of the correlation over time.

changing during the period of our data. Indeed, we findmajor changes over time for academic research productiv-ity. A paired difference test between academic research inthe first and last years of our data shows that the change issignificantly different from zero (p < .01). In addition, asexpected, we find that the variance of academic research is

significantly higher than the variance in performance per-ceptions for each type of constituent (p < .01).

Impact of Academic Research on Perceptions ofMultiple Constituents

In Table 4, we report the effect of academic research on theperformance perceptions of three constituents of businessschoolsacademics, recruiters, and applicants. Both thelagged perceptions and the lagged academic researchparameters are significant. As we discussed previously, thisindicates carryover; that is, a change in academic researchtends to have persistent effects on MBA program perfor-mance over time (Jacobson 1990). In other words, the per-ceptions of an MBA program in a particular year are depen-dent not only on research during that year but also onresearch in previous years. Although these carryover effectspersist over an infinite time horizon, the empirical questionsare as follows: (1) What are the short-term and long-termeffect sizes? and (2) What are the durations of these effects?Therefore, we compute effect sizes and carryover durationsfor academic research on the performance perceptions ofacademics, recruiters, and applicants.

We find a significant short-term effect of academicresearch on academics perceptions (p < .01), meaning that

more research is associated with improved perceptions (i.e.,lower rank). As we expected, academic research does nothave a significant short-term effect on recruiters and appli-cants perceptions. In contrast, over the long run, we findsignificant effects of academic research on academics per-ceptions (p < .01) and a marginally significant effect on

recruiters and applicants perceptions (p < .1). Althoughthese results may support the belief (or at least the hope) ofmany business school faculty members, they provide thefirst empirical evidence of these important effects of acade-mic research. Moreover, they help refute the conjectures ofsome authors (i.e., Bennis and OToole 2005; Pfeffer andFong 2002) that academic research is a disservice to a busi-ness schools nonfaculty constituents. Overall, these resultssupport H1 and H2.

H3 predicts that the carryover duration will be the short-est for academics, and we find support for this hypothesis.In addition, we find that applicants perceptions have thelongest carryover duration.

Research productivity during any single year for the 57business schools in our database ranges from the equivalentof 1 to 31 single-author articles (average = 8.3). We findthat an increase of one single-author article has a long-termimpact of improving academics perceptions by one-third ofa rank and recruiters perceptions by one-sixth of a rank.Similarly, one additional article improves applicants per-ceptions in the long run, as reflected by a decline in accep-tance rate of .38%. However, on average, across the threetypes of constituents, it takes more than four years to realizethese benefits of academic research. We believe that these


11/20


TABLE 4Effect of Academic Research on Multiple Constituents Perceptions

Academics Recruiters ApplicantsPerceptions Ranka Perceptions Ranka Acceptance Rateb

Intercept 17.46*** 15.59*** 15.12***Lag dependent variable .42*** .44*** .63***Academic research .08** .02 .07Lag academic research .12*** .11** .07*University endowmentc .12*** .22*** .06***

R2 (%) 34.3 24.9 39.7Number of observations 799 799 722

Unit Ordinal Ordinal Percentage

Short-term effect of a single-author articled .12*** Not significant Not significantLong-term effect of a single-author articled .34*** .16* .38*90% carryover duration of a single-author articled 3.35 years 4.12 years 5.54 years

*p .10.**p .05.***p .01.aMeasured in ordinal rank; perception improves when rank decreases.bMeasured as a percentage of applicants accepted; perception improves when acceptance rate declines.cMeasured in units of hundreds of millions of dollars.dBased on Equation 2.

results are highly important. They demonstrate the positive

impact of academic research on several constituents of busi-

ness schools outside their own faculty members.

Impact of Academic Research on Education

Performance of Business Schools

In Table 5, we report the impact of academic research on

education performance of business schools. As we dis-

cussed previously, we examine three related performance

measureseffectiveness, efficiency, and overall perfor-

mance. Next, we discuss the results of estimating Equation

4. Because effectiveness is the amount of starting salary in

excess of that predicted by the GMAT score, we control for

the monetary inputs (i.e., tuition and endowment). Simi-

larly, because efficiency is the amount of starting salary in

excess of that predicted by the tuition and endowment, we

TABLE 5

Effect of Academic Research on Education Performance

Variable Equation 4 Effectivenessa Efficiencyb Overall Performancec

Intercept b0 6630.86** 23,283.40* 297.79Lag education performance b1 .32*** .38*** .36***Academic research b2 36.61 32.89 45.19Lag academic research b3 110.07** 121.38** 137.32**GMATd 36.74*University endowmente 85.59***Yearly tuitionf 227.88**

Number of observations 799 749 749

Long-term effect of one single-author article for abusiness school (in starting salary $/student)g 215.71** 248.82** 285.17**

90% carryover duration of one single-author articlefor a business school (in years)g 2.86 3.23 3.08

*p .10.**p .05.***p .01.aMeasured as a students starting annual salary in excess of that predicted by the students GMAT score, in 1998 dollars.bMeasured as a students starting annual salary in excess of that predicted by tuition and university endowment, in 1998 dollars.cMeasured as a students starting annual salary in excess of that predicted by the students GMAT score and tuition, as well as universityendowment, in 1998 dollars.

dLagged by two years to reflect the GMAT score of the same group of students with the expected salary.eMeasured in hundreds of millions of dollars.fMeasured in thousands of dollars.gBased on Equation 2.


12/20


11This effect is positive for an increase in the number of articlesand negative for a decrease in the number of articles.

control for the nonmonetary input (i.e., graduates GMATscores). Because our overall performance measure is theamount of starting salary in excess of that predicted byGMAT score, tuition, and endowment, we do not control forthese inputs.

For all three of our measures of education performance,we find a significant effect of lagged academic research.The parameters for academic research, lagged academicresearch, and the lagged dependent variables depict a state-dependent carryover process. We compute the effect size

and the 90% carryover duration according to these parame-ter values. We find that academic research has significantlong-term effects on all three measures of education perfor-mance (p < .05). For effectiveness, we estimate the long-term effect of one single-author article to be more than$200 (in 1998 dollars) on the starting annual salary of agraduating student.11 Similarly, for efficiency, the long-termeffect is approximately $250, and for overall performance,

it is closer to $300. Across all three models, the carryoverduration of the long-term effect of academic research isapproximately three years. Overall, the consistency inresults across our three measures of education performanceprovides strong support for H4.

Impact of Reputation

In Table 6, we report the parameters of our hierarchicalmodel equations (Equations 5a and 5b). We find that the

long-term effects of academic research across all three mea-sures of education performance are significant for the busi-ness schools in high-reputation universities. We computethe function of parameters in Equation 6 to examinewhether these effects are larger than those of businessschools in other universities (i.e., those that are not in ahigh-reputation, top-20 university). For effectiveness, wefind significant support for the hypothesized effect of repu-tation (p < .05). For the overall performance measure, wefind marginally significant support (p < .10). For efficiency,we find a directionally consistent result, though it is not sig-nificant. In addition, based on our overall measure of educa-tion performance, the long-term benefit of an additionalsingle-author article for business schools in high-reputation

TABLE 6Reputations Influence in Effect of Academic Research on Education Performance

Equations OverallVariable 5a and 5b Effectivenessa Efficiencyb Performancec

Intercept b0 6473.15* 25,167.5* 308.95Lag education performance b1 .34*** .39*** .36***Academic research b2 69.10 35.52 43.49*Lag academic research b3 113.43** 145.99** 119.05*Improvement 10 .03 .04 .06*

20 89.21* 108.70 62.35

30 123.14** 64.74** 87.96*Improvement high reputation 11 .11** .09* .08**21 129.14*** 145.49*** 116.29**31 36.04* 78.13* 31.85*

GMATd 39.60*University endowmente 82.17***Yearly tuitionf 231.71**

Number of observations 799 749 749

Long-term effect of one single- High reputation 418.32** 310.47* 413.02**author article (in starting salary + improvement$/student) Others + improvement 343.59** 241.32 324.40*

90% carryover duration of one High reputation 2.49 2.89 2.88single-author article (in years)g + improvement

Others + improvement 3.37 4.04 3.72

*p .10.**p .05.***p .01.aMeasured as a students starting annual salary in excess of that predicted by the students GMAT score, in 1998 dollars.bMeasured as a students starting annual salary in excess of that predicted by tuition and university endowment, in 1998 dollars.cMeasured as a students starting annual salary in excess of that predicted by the students GMAT score and tuition, as well as universityendowment, in 1998 dollars.dLagged by two years to reflect the GMAT score of the same group of students with the expected salary.eMeasured in hundreds of millions of dollars.fMeasured in thousands of dollars.gBased on Equation 2.


13/20


universities is more than $380 per graduating student. For asimilar increase, the benefit obtained by the other businessschools in our data is approximately $300 per graduatingstudent. Therefore, the benefits of research at businessschools in high-reputation universities are more than 25%higher than those at other business schools. These resultssupport H5.

Next, we compute the carryover duration of an increasein academic research for high-reputation business schools.We find that for these schools, 90% of the benefits tend to

be realized over an average of 2.75 years (based on an aver-age of the three education performance measures). Forother business schools, the same benefit takes approxi-mately 3.7 years (i.e., 35% longer) to be realized. Althoughwe cannot directly test for the significance of carryoverduration, we can examine the parameters used in computingthis duration. Specifically, we find two results that drive thisdifference in carryover duration. First, 11 is significant andnegative, thus reducing the coefficient of the lagged depen-dent variable. This result means that business schools inhigh-reputation universities have a lower proportion of car-ryover. Second, 21 is significant and positive, whereas 31is significant and negative. Furthermore, 21 is related to the

coefficient of current academic research, whereas 31 isrelated to the coefficient of lagged academic research.These three parameters all contribute to business schools inhigh-reputation universities having significantly largershort-term returns for increases in academic research.Therefore, the carryover duration is shorter, in support ofH6. Overall, this important asymmetry for reputation mayprovide a reason that high-reputation business schools tendto maintain their reputations over time.

Trends in the Effect Size and Effect Duration ofAcademic Research

We now explore whether the impact of academic research

has changed over time across our six dependent measuresencompassing constituents perceptions and businessschools education performance. Because we need a fewyears worth of data to be able to estimate our models, weadopt a moving window of five years to estimate effectsizes and durations. Table 7 reports the results of ourestimations. We find two significant trends. First, the effectsize of academic research has a significant, negative trendfor recruiters perception rank (p < .05), which meansthat recruiters are increasingly sensitive to academicresearch. This evidence runs counter to recent calls forde-emphasizing research in business schools. Second, wefind that the carryover duration for academic research onall three measures of education performance is significantlydecreasing over time (p < .05). Similarly, we find margin-ally significant increases in effect sizes for education per-formance (i.e., student salaries) over time.

These findings indicate that the educational value ofacademic research is being realized more quickly and morestrongly. Notably, these three trends are associated withsome media (e.g., BusinessWeek, Financial Times) thatincorporate a research component into their MBA programrankings. In addition, as business schools have gained vis-ibility, innovative research has met with unprecedented

media interest (BusinessWeek2007). Surprisingly, we findthat the effect size of academic research has a marginallysignificant, positive trend for academics perception rank(p < .1), which means that academics are becoming some-what less sensitive to academic research.

Robustness of Our Results

Robustness to new data and changes in measurement.First, our three measures of education performance arederived from the differentials between output and inputs.We believe that this approach reflects the overall educa-tional environment within a business school and its ultimateimpact on students salaries. However, other researchershave used students instructor ratings as a more direct mea-sure of education performance. Although these measuresmay also reflect the popularity of instructors, studentsexpected grade, and class size, we still want to evaluate ourresults with this alternative measure of education perfor-mance. We were able to obtain these ratings at the courselevel during the 20012006 period for two business schoolsin our data set. We use these ratings in two ways. First, wecompute the average instructor rating for each year and thencalculate the correlations between these average ratings and

the three measures of education performance. We find thatthe correlations range between .78 and .80 for one schooland between .64 and .74 for the other school. These highcorrelations reflect our belief that there is some overlap inthese measures, though they seem to be capturing differentfactors as well. Second, we model the short-term and long-term effects of academic research on students instructorratings across departments within each business school. Weaggregate instructor ratings and academic publications intofive departments: accounting, finance, information systems/operations, management, and marketing. Thus, for bothbusiness schools over six years, we have data on fivedepartments for academic research and students instructor

ratings. On the basis of these data, we use Equation 2 toestimate the short-term and long-term relationships betweenacademic research and students instructor ratings acrossdepartments within each business school. We do not use anyof the control variables used in our previous models,because endowment, tuition, and GMAT scores areassumed to be the same across departments within eachbusiness school. In Table 8, we present the parameter esti-mates. For both business schools, we find that there are sig-nificant long-term effects of academic research (p < .05).The carryover durations of 2.2 years and 4.5 years are simi-lar to the carryover durations for our original three mea-sures of education performance. Overall, these results helpvalidate our measures of education performance. Moreimportant, they confirm our finding that academic researchhas positive long-term effects on education performance inbusiness schools.

Second, we evaluate an alternative to GMAT scores formeasuring incoming student quality for our models of effi-ciency and overall education performance. If admissionsdirectors consider additional factors beyond GMAT scores,acceptance rate may provide a more general measure of stu-dent quality than GMAT scores. Therefore, we use accep-tance rate (lagged by two years), instead of GMAT scores,


14/20


Long-Term Effect Size of Academic Researcha

Education Performance

AcademicsPerceptions

Rank (Ordinal)

RecruitersPerceptions

Rank (Ordinal)

ApplicantsAcceptance

Rate (%)

Effectiveness(in 1998Dollars)

Efficiency(in 1998Dollars)

OverallPerformance

(in 1998 Dollars)

20022006 .13 .58 .26 297.06 225.19 241.03

20002004 .28 .66 .15 391.23 338.81 356.0319982002 .22 .50 .65 450.65 360.53 396.0519962000 .39 .15 .50 104.87 150.70 415.4819941998 .25 .03 .09 261.54 244.78 143.2719921996 .33 .05 .12 69.35 65.69 235.0319901994 .35 .28 .40 88.91 102.39 197.11Linear trendb Positive*

(p< .09)Negative**(p< .05)

Not significant Positive*(p< .06)

Positive*(p< .08)

Not significant

Long-Term Effect Duration of Academic Research

AcademicsPerceptionsRank (Years)

RecruitersPerceptionsRank (Years)

ApplicantsAcceptanceRate (Years)

Education Performance

Effectiveness(Years)

Efficiency(Years)

OverallPerformance

(Years)

20022006 2.47 2.61 3.49 2.78 2.69 2.6020002004 3.31 3.13 4.40 2.97 3.25 3.4719982002 2.40 5.07 7.21 3.73 4.11 2.8719962000 2.66 3.03 3.88 2.93 2.97 3.6019941998 2.48 5.09 3.50 3.71 3.90 4.8619921996 4.46 N.A.c 3.49 4.24 4.41 5.7819901994 4.89 5.73 4.12 4.03 4.19 4.99Linear trend Negative*

(p< .07)Negative*(p< .06)

Not significant Negative**(p< .02)

Negative**(p< .05)

Negative***(p< .01)

TABLE 7

Trends in the Long-Term Effect Size and Effect Duration of Academic Research

*p .10.**p .05.***p .01.aFor academicsperceptions rank, recruiters perceptions rank, and applicants acceptance rate, a negative sign in long-term effect size denotesan improvement in rank or a decrease in acceptance rate for an increase in academic research.

bFor academics perceptions rank, recruiters perceptions rank, and applicants acceptance rate, a positive trend in long-term effect size

denotes a decrease in effect size over time, and a negative trend denotes an increase in effect size over time. For education performance,positive trends in effect size denote increases in effect size over time, and negative trends denote decreases in effect size over time.

cDuration is indeterminate in this case because the parameter of academic research is positive and larger than the negative lagged academicresearch parameter.

to calculate educational effectiveness. Then, we model theimpact of academic research on this alternative measure ofeducational effectiveness. The parameter estimates are notsignificantly different from our GMAT-based results, andthe results on all hypotheses are the same.

Third, we consider graduates average starting salary asthe output measure of education performance, without alsofactoring in the percentage employed three months after

graduation. Again, the results on all hypotheses are thesame.

Fourth, our measure of academic research does notaccount for the size of the faculty in a school. Unfortu-nately, we cannot obtain the faculty size of the 57 businessschools over the complete period. Instead, we use threeapproaches. First, we obtain publicly available data col-lected by Washington State University researchers on thenumber of publications per faculty member in the market-ing departments of business schools (see www.vancouver.

wsu.edu/fac/cote_joe/pubstudy/productivity.htm). Thesedata are based on publications in Journal of ConsumerResearch, Journal of Marketing, and Journal of MarketingResearch between 1984 and 1999. The correlation betweenthe total number of publications and the number of publica-tions per faculty is .83 (p < .001). Although these results arerelated only to marketing faculties, there is no reason theywould be different for other disciplines. Second, we use stu-

dent enrollment as a surrogate for faculty size. We find thatthis proxy measure is highly correlated (.887) with univer-sity endowment (a variable included in our model). Third,we divide our measure of academic research by studentenrollment to obtain a measure of academic research perstudent. We rerun our analyses using this measure and,again, find support for all our hypotheses.

Fifth, we evaluate whether using business schoolendowment, instead of university endowment, affects ourresults. Although business school endowment data are more


15/20


Variables

Department-Level

InstructorRating of

B-School Xa

Department-Level

InstructorRating of

B-School Ya

Intercept 3.85*** 2.13***Lagged rating 0.18*** 0.48***Academic research 0.0086 .0095Lagged academic

research 0.021*** 0.026***R2 (%) 27.9 35.4

Unit17 Rating

Scale15 Rating

Scale

Short-term effectof one single-author articleb n.s. n.s.

Long-term effectof one single-author articleb .036** .032**

90% carryoverduration of onesingle-authorarticleb 2.18 years 4.50 years

TABLE 8

Effect of Academic Research on StudentsInstructor Ratings in Two Business Schoolsa

*p .10.**p .05.***p .01.aInstructor rating measured on a 17 scale for departments of Busi-ness School X and on a 15 scale for departments of BusinessSchool Y.

bBased on Equation 2.Notes: n.s. = not significant.

difficult to obtain, we were able to find 2007 data for 46 ofthe 57 business schools. Using this alternative measure of

wealth, we still find support for three of our four hypotheseson the short-term and long-term effects. Only the long-termeffect on applicants perceptions changes from marginallysignificant (p < .1) to not significant.

Sixth, we examine the possibility that there are differenteffects of research on education at top-tier and second-tierbusiness schools. We split our data at the median of theschools overall USNrankings, averaged over all years. Wererun the model on the effects of academic research onoverall education performance separately for top-tier andsecond-tier schools. In both samples, we find significantlong-term effects of research on education performance.

Seventh, we test the sensitivity of our results on reputa-tion effects by considering alternative cutoffs in our catego-rization of high-reputation universities. Specifically, we cat-egorize either top-15 universities or top-25 universities ashigh-reputation universities, instead of our previous top-20categorization. None of the parameters in Tables 5 or 6change by more than one standard error as a result of thedifferent cutoffs, and the results on all hypotheses remainthe same.

Eighth, we consider the effects of public versus privateuniversities by including a categorical variable in the mod-

els. We do not find any significant effects for this categori-cal variable.

Robustness to changes in model specification and esti-

mation. First, we consider our model specification and theimportant issue of endogeneity. There are two possiblesources of endogeneityunobserved variables and otherspecification errors (e.g., reverse causality). Unobservedvariables, such as students work experience, salary beforematriculation, organization culture, or even random luck,could play a significant role in driving perceptions and per-formance. Although no method other than controlled exper-iments can truly recover causality from observational data,the time-series nature of our data enables us to examine theveracity of alternative explanations. To account for unob-served variables, we use lagged instruments for our depen-dent variablesthat is, Ri(t 1) and Ri(t 2) in place of Ritand Ri(t 1). Thus, unobserved variables (at least those thatare not autocorrelated) cannot influence the associationbetween these lagged instruments and the dependentvariables in our model, because the future cannot cause thepast (Geweke 1984; Granger 1969; Jacobson 1990). There-fore, we modify Equation 2 as follows:

Pijt

= b0ij

+ b1ij

Pij(t 1)

+ b2ij

Ri t 1

+ b3ij

Ri(t 2)

+ijt

.

On the basis of the results of this analysis, we continue tofind support for all our hypotheses. Furthermore, the long-term effect sizes remain within one standard error of ouroriginal model estimates.

Second, we conduct five model specification tests. First,we check for unit roots using an augmented DickeyFuller(ADF) test (Dekimpe and Hanssens 1995). Dekimpe andHanssens (2003, p. 24) show that the appropriate modelspecification is contingent on the results of the ADF test. Inparticular, an autoregressive model at the absolute level ofthe variables is preferred when these variables are stable,whereas a difference model is preferred when these

variables are evolving. We include two lags in the ADFmodel based on the Akaike information criterion (Nijs et al.2001). For each of the seven variables (academic research,three constituents perceptions, and three measures of edu-cation performance) and across all business schools, wefind that the unit root null is rejected (for six of the sevenvariables, p < .01; for the other, p < .05). We also do notfind any significant trend in these variables. These resultsindicate that our variables are stable and that our use of theabsolute level of the variables is appropriate (Dekimpe andHanssens 2003). Second, on the basis of the doubleprewhitening method, we find that a higher level of cross-correlation exists between the residuals of all six perfor-mance variables and academic research at a positive lag (fordetails, see Hanssens, Parsons, and Schultz 2001, pp. 31315). Third, we use regression-based Granger- and Sims-causality tests. We find that all six relationships with acade-mic research (i.e., academics, recruiters, and applicantsperceptions, as well as effectiveness, efficiency, and overalleducation performance) exhibit causality in the proposeddirection. Only academics perceptions also exhibit signifi-cant causality in the reverse direction (p < .05). Fourth, werun simultaneous equation models as follows:


16/20


Pijt = b0j + b1jPij(t 1) + b2jRi(t 1) + b3jRi(t 2) + ijt, and

Rit = c0j + c1jRi(t 1) + c2jPij(t 1) + c3jPij(t 2) + it.

The objective is to evaluate how many of the significantrelationships in our proposed causal direction are also sig-nificant in the reverse causal direction. We find that onlythree of the seven significant results in Table 4 are signifi-cant in the reverse direction. Moreover, all these relation-ships have t-statistics that are much smaller than those inthe proposed direction. Although these results do not rule

out reverse causality, they provide some support for ourproposed causal structure. Fifth, because the number oftime series is limited to 17 years, the asymptotic consis-tency of the parameters can be questioned. Anderson andHsiao (1981) propose a simple consistent estimator for lim-ited time series based on an instrumental variable proce-dure. Thus, we use a first differences model:

(Pijt Pijt 1) = b1ij(Pijt 1 Pijt 2) + b2ij(Rijt Rijt 1)

+ b3ij(Rijt 1 Rijt 2) + (ijt ijt 1).

Because the error term (ijt ijt 1) is correlated with(Pijt 1 Pijt 2), Anderson and Hsiao recommend Pijt 2 asan instrumental variable for (Pijt 1 Pijt 2). With this alter-native specification, we estimate our parameters and usethese estimates to compute the short-term and long-termeffects. For academics perceptions, recruiters perceptions,and overall education performance, the computed long-termeffects remain significant and within one standard error ofour original estimates. The long-term effect is not signifi-cant for students perceptions (i.e., acceptance rate).

Third, we consider heteroskedasticity in our estimatederror. We use jackknifing techniques to identify MBA pro-grams with large average residuals. We isolate these pro-grams sequentially and repeat the GoldfeldQuandt testuntil the homoskedastic null is accepted (Sayrs 1989).Then, we reestimate Equation 3 without six such MBA pro-

grams. Overall, these estimates are not significantly differ-ent from the base model estimates.

Fourth, although the parameter estimates of Equations 3and 4 do not indicate serial correlation, we use a first-orderautocorrelated error structure to check the robustness of ourresults. For this model, we reestimate these models byadding the condition it = i i(t 1) + it. We use the Parksmethod to estimate this models parameters and the covari-ance matrix using a two-stage generalized least squares pro-cedure. The results of this model are virtually identical toour original estimates.

DiscussionFaculty members at most leading business schools devote asubstantial amount of their time to academic research. Sur-prisingly, however, there is little empirical support for thebenefits of these endeavors. Even more surprisingly, theresearch that exists questions the value of academicresearch in fulfilling a business schools broad educationalmission. This situation has provided an opening for criticsof research to question its usefulness in todays businessschools. To address this lack of empirical research and

begin to address some of the limitations in existingresearch, we evaluate the impact of academic research onmultiple measures of business school performance. On thebasis of data from 57 business schools over 18 years, wefind the following:

Academic research has positive long-term effects on the per-ceptions of business schools by other academics, recruiters,and applicants. Among academics, some of these effects arealso realized in the short run.

90% of the long-term effects on perceptions are realized over

three to six years.The effect sizes are large and meaningful. When a businessschool has a persistent increase of three single-author articlesper year, its ranking among other academics improves by oneplace, and its student acceptance rate declines by 1%.

Academic research has a positive long-term effect on threemeasures of education performance and, at least at two busi-ness schools, on students instructor ratings.

A persistent increase of one single-author article per year isassociated with an increase in graduates average startingsalary of approximately $250 (in 1998 dollars). This effect isapproximately 25% larger for business schools in higher-reputation universities in our sample.

More recently, academic research has had a stronger effect on

recruiters perceptions and shorter carryover duration on edu-cation performance.

Implications

Our findings have two general implications for all types oforganizations and eight additional implications for businessschools. First, our study demonstrates the importance ofevaluating both short-term and long-term effects of an orga-nizations activities. Previous research has focused on thecontemporaneous relationship between research and educa-tion, but we find that the predominant impact of researchoccurs over the long run. Second, our study demonstratesthe importance of considering various performance objec-

tives held by multiple constituents. The impact of activitiesmay differ across these objectives, and it is important tounderstand their effect sizes to allocate resources more effi-ciently. This effort is particularly important in nonprofitorganizations because activities cannot be evaluated againsta common profit metric.

With regard to business schools, our first implication isthat the roles of knowledge exploration (research) andknowledge exploitation (education) in business schools arenot competing activities. Our results show that explorationdirectly benefits exploitation in the short run with some ofour performance measures and in the long run with all ofour performance measures. Thus, business schools empha-

sizing research are not favoring faculty at the expense ofother constituents, as recent critics have charged. Second,business school administrators, who might previously haveviewed research as competing with other business schoolactivities, should now allocate more resources to research inlight of its positive effects on the perceptions of importantconstituents and education performance. We hope that ourstudy can begin to put an end to the belief that emphasizingacademic research is not good for business schools broadereducational mission (Bennis and OToole 2005; Pfeffer and


17/20


Fong 2002). Such a revised belief should also lead to moregovernment funding of business school research because ofits ability to contribute to the general public good by edu-cating students and placing them in positions more valuedby firms. Third, investments in academic research willbenefit business schools in several ways. They will increasethe overall performance of MBA programs, as measured byBusinessWeek, USN, Financial Times, and other publica-tions. Possible mechanisms for this overall improvementare evident in how academic research improves reputations

among academics, recruiters, and applicants. Fourth, busi-ness school administrators must make a long-term commit-ment to improving research to benefit from such a strategy.Todays increased investment in research will likely gener-ate more journal submissions in two to three years. Then, itwill take another two to three years for some of these papersto be published. Therefore, our finding of a three- to six-year carryover should be roughly doubled to realize the fullreturn on increased investments in research. As a result,business schools may be wise to increase the tenure of theirdeans or to provide incentives based on long-term perfor-mance. These steps will ensure that a strategy of emphasiz-ing research is more likely to reward the dean who invests

in that strategy rather than his or her successor. Currently,the lack of knowledge about long-term effects means thatpoor performance in rankings is associated with a signifi-cant increase in the [immediate] likelihood of a dean leav-ing (Fee, Hadlock, and Pierce 2005, p. 143). Fifth,although the effects of academic research appear to be morebeneficial for MBA programs at higher-reputation universi-ties, other schools derive significant benefits as well. Sixth,business school administrators may want to undertakeefforts to shorten the duration of the carryover effects bymore actively promoting faculty members research to stu-dents, recruiters, and reporters. Students could be educatedabout the importance of research when selecting which

school to attend because research will affect the quality oftheir educational experience and the future ranking of theirschool. Seventh, business schools may want to increasetheir recruitment of faculty members with articles in

advanced stages of review to capture the benefits of theirpublications. Eighth, media rankings of MBA programsshould incorporate an increased weight for academicresearch because of its relationship to multiple constituentsand the objectives of business schools. Moreover, such ameasure of academic research should include each disci-plines core academic journals, in addition to morepractitioner-oriented journals and books.

Directions for Further Research

Our studys findings and its limitations provide severalopportunities for further research. First, researchers shouldexpand the investigation of short-term and long-term effectsacross a broad range of organizational activities. In somecases, effects may be positive in the short run but not sig-nificant, or may even be negative, in the long run. By study-ing a variety of contexts, researchers may be able touncover some generalizations about the lag structuresbetween organizational activities and performance. Second,future studies on academic research could use panel data atthe level of individual professors to examine the long-termeffects of research on student ratings. In particular, it would

be worthwhile to examine differences among disciplines,stages of careers, and a host of other individual characteris-tics. Third, future studies on business schools and othereducational institutions could include lower-reputationinstitutions. Although we found an important effect of repu-tation, the role of academic research may be different atschools ranked outside the top 50 or at colleges that viewtheir mission as primarily one of teaching. Fourth,researchers should try to replicate our findings on under-graduate business school programs. Finally, more fine-grained studies could lead to a better understanding of theprocess by which academic research has a positive impacton constituents perceptions and education performance.

For example, do certain types of research have more of animpact on certain objectives? All these directions for furtherresearch should help expand knowledge of the importantrole of academic research in education.

REFERENCESAMA Task Force on the Development of Marketing Thought

(1988), Developing, Disseminating, and Utilizing MarketingKnowledge,Journal of Marketing, 52 (October), 125.

Amemiya, Takeshi (1985), Advanced Econometrics. Cambridge,MA: Harvard University Press.

Anderson, Eugene W. and Linda Court Salisbury (2003), TheFormation of Market-Level Expectations and Its Covariates,

Journal of Consumer Research, 30 (1), 11524.Anderson, T.W. and Cheng Hsiao (1981), Estimation of DynamicModels with Error Components,Journal of the American Sta-tistical Association, 76 (September), 598606.

Arrow, Kenneth J., Hollis B. Chenery, Bagicha S. Minhas, andRobert M. Solow (1961), Capital-Labor Substitution and Eco-nomic Efficiency,Review of Economics and Statistics, 43 (3),22550.

Banker, Rajiv D., Gordon B. Davis, and Sandra S. Slaughter(1998), Software Development Practices, Software Complex-ity, and Software Maintenance Performance: A Field Study,

Management Science, 44 (4), 43350.

Bartelsman, Eric J. and Phoebus J. Dhrymes (1998), Productivity

Dynamics: U.S. Manufacturing Plants, 19721986,Journal of

Productivity Analysis, 9 (1), 534.

Bennis, Warren G. and James OToole (2005), How Business

Schools Lost Their Way,Harvard Business Review, 83 (May),

96104.

Boulding, William, Ajay Kalra, and Richard Staelin (1999), The

Quality Double Whammy,Marketing Science, 18 (4), 46384.

Brown, Stephen J. and Jerold B. Warner (1985), Using Daily

Stock Returns: The Case of Event Studies, Journal of Finan-

cial Economics, 14 (1), 331.

Bryk, Anthony S. and Stephen W. Raudenbush (1992), Hierarchi-

cal Linear Models: Applications and Data Analysis Methods.

Newbury Park, CA: Sage Publications.

BusinessWeek (2007), Powerful Profs, (August 22), (accessed

May 14, 2008), [http://www.businessweek.com/bschools/

content/aug2007/bs20070821_430502.htm].


18/20


Clarke, D.G. (1976), Econometric Measurement of the Durationof Advertising Effect on Sales, Journal of Marketing

Research, 13 (November), 34557.Conover, W.J. and R.L. Iman (1981), Rank Transformations as a

Bridge Between Parametric and Nonparametric Statistics, The American Statistician, 35 (3), 12429.

Cowan, J. (1985), Effectiveness and Efficiency in Higher Educa-tion,Higher Education, 14 (3), 23539.

DAveni, Richard A. (1996), A Multiple Constituency, Status-Based Approach to Interorganizational Mobility of Faculty andInput-Output Competition Among Top Business Schools,

Organization Science, 7 (2), 16689.Dekimpe, Marnik G. and Dominique M. Hanssens (1995), The

Persistence of Marketing Effects on Sales,Marketing Science,14 (1), 121.

and (1999), Sustained Spending and PersistentResponse: A New Look at Long-Term Marketing Profitability,

Journal of Marketing Research, 36 (November), 397412. and (2003), Persistence Modeling for Assessing

Marketing Strategy Performance, ERIM Report SeriesResearch in Management, ERS-2003-088-MKT.

Epley, D.R. (1997), A Note on the Optimal Selection and Weight-ing on Comparable Properties, Journal of Real Estate

Research, 14 (2), 17582.Fairweather, J.S. (2002), The Mythologies of Faculty Productiv-

ity,Journal of Higher Education, 73 (1), 2648.

Fama, Eugene F. and Kenneth R. French (1992), The Cross-Section of Expected Stock Returns, Journal of Finance, 47(2), 42765.

Fee, C.E., C.J. Hadlock, and J.R. Pierce (2005), Business SchoolRankings and Business School Deans: A Study of Non-ProfitGovernance, Financial Management, 34 (1), 14356.

Feldman, K. (1987), Research Productivity and AcademicAccomplishment of College Teachers as Related to TheirInstructional Effectiveness: A Review and Exploration,

Research in Higher Education, 26 (3), 22798.Gallagher, Katherine B. and Charles B. Weinberg (1991), Coping

with Success: New Challenges for Non-Profit Marketing,Sloan Management Review, 33 (1), 2742.

Geers, A.L. and G.D. Lassiter (1999), Affective Expectations andInformation Gain: Evidence for Assimilation and Contrast

Effects in Affective Experience, Journal of ExperimentalSocial Psychology, 35 (4), 394413.Geweke, J. (1984), Inference and Causality in Economic Time

Series Models, in Handbook of Econometrics, Vol. 2, Z.Griliches and M. Intriligator, eds. New York: North-Holland,11021144.

Golder, Peter N. (2000), Historical Method in MarketingResearch with New Eviden

Documents

Does Academic Research Help or Hurt MBA Programs