Agency Theory and Variable Pay Compensation Strategies.

© Academy of Management Journal1996, Vol. 39, No. 3, 751-767.

AGENCY THEORY AND VARIABLE PAYCOMPENSATION STRATEGIES

LINDA K, STROHLoyola University Chicago

JEANNE M. BRETTNorthwestern UniversityJOSEPH P. BAUMANN

Northwestern UniversityANNE H. REILLY

Loyola University Chicago

This study used a sample of middle-level managers to investigate theeffects of organization-level agency-theory-based variables on the pro-portion of variable compensation that managers receive. Level of taskprogrammability was associated with an increased use of variable pay,and long-term relationships between an agent and principal were associ-ated with decreased use. Results supported the classical organization-theory prediction that under higher risk, organizations use higher pro-portions of variable pay; but results question agency theory's ability topredict compensation strategy for middle-level managers in the high-risk situation.

Agency theory is an outgrowth of research on risk sharing conductedby economists (Boyd, 1994; Eisenhardt, 1988; Gomez-Mejia & Balkin, 1992a;Lambert, Larcker, & Weigelt, 1993). According to agency theory, each firmconsists of a principal (in our study, an organization) and an agent (in ourstudy, a manager). The assumptions of agency theory are that agents aremotivated by self-interest, are rational actors, and are risk-averse. Therefore,principals can motivate agents by controlling their incentives (Amernic,1984; Eisenhardt, 1989; Gomez-Mejia & Balkin, 1992a, 1992b). An agencydilemma occurs, however, when a principal is unable to adequately monitoror assess an agent's behavior (Amernic, 1984; Lambert et al., 1993). Thissituation results when the agent's task is less programmable (Eisenhardt,1989), when accomplishing the task entails risks, or when the goals of theprincipal and agent are in conflict. In this study, we developed and testedhypotheses about the relationship between the use of variable pay incentivesfor managers and three variables that are prominent in classical agency the-ory: task programmability (when it is low, an agent's behavior is difficult tomonitor); risk (which occurs when an organization is undergoing rapidchange); and the length of the agent-principal relationship (when it is long-lasting, the goals of the principal and the agent are aligned) (cf. Alchian &Demsetz, 1972; Eisenhardt, 1989).

751

752 Academy of Management Journal June

Eisenhardt (1989) suggested that it is appropriate to use agency theoryto examine internal compensation schemes. However, most agency theoryresearch has focused on top executive compensation (Balkin & Gomez-Mejia,1990; Beatty & Zajac, 1994; Gerhart & Milkovich, 1990; Westphal & Zajac,1994), and there has been little research testing agency theory predictionsfor middle managers. One of this study's contributions is its application ofagency theory constructs to the variable pay experience of middle managers.

Organizations bave considerable discretion in choosing compensationstrategies for tbeir managers (Balkin & Gomez-Mejia, 1987; Gerhart & Milko-vich, 1990; Gomez-Mejia & Welbourne, 1988; Lawler, 1981, 1990). There areat least two options: fixed pay (salary) and variable pay (bonus, incentiveplans, profit-sbaring, ESOPs). Researcb bas shown tbat sucb strategies tendto be aligned witb (1) tbe cbaracteristics of industries, sucb as tbeir tecbnolog-ical focus and variation in product demand (Baker, Jensen, & Murpby, 1988;Gerhart & Milkovicb, 1990; Maboney, 1979); (2) tbe characteristics of organ-izations themselves (Mellow, 1982), such as the extent and process of theirdiversification, tbeir strategic business unit (SBU) level strategy, and theirlife cycle stage (cf. Gomez-Mejia & Balkin, 1992a); and (3) the characteris-tics of employees, sucb as tbeir education and skill levels (Becker, 1975).Industry-level and organizational factors, bowever, are distal and may notbe particularly salient to managers. Gerbart and Milkovicb (1990) pointedout tbat otber organizational factors may influence compensation contractsfor managers. However, tbey did not identify wbat tbose factors migbt be.

In tbis article, we draw upon agency tbeory to develop bypotheses aboutorganizational cbaracteristics proximal to tbe day-to-day jobs of managersand variable pay compensation scbemes. We begin by discussing tbe agencytbeory constructs represented by our exogenous variables; we tben developbypotbeses regarding tbeir relationsbips witb tbe proportion of variable payreceived by an organization's managers. We also propose a competing riskpremium bypotbesis related to total casb compensation.

THEORY AND HYPOTHESES

Task Programmability

Agency tbeorists propose tbat tbe fundamental goal of firms is to mini-mize costs and maximize efficiency. As it relates to compensation strategy,agency tbeory suggests tbat organizations cboose between fixed and variablepay by determining bow easy it is to monitor job performance (bebaviors).One factor related to the ability to monitor performance is task programma-bility.

A programmable task is one wbose requisite bebaviors can be preciselydefined (Eisenbardt, 1989). According to agency theory, task programmabilitywill be positively related to tbe use of bebavior-based compensation contracts(fixed salary) and negatively related to tbe use of outcome-based contracts(variable pay). Tbis is because programmable tasks allow the principals (inthis case, organizations) to specify the behaviors that tbe agents (in tbis

1996 Stroh, Brett, Baumann, and Reilly 753

case, managers) need to perform. Gomez-Mejia and Balkin (1992b) used theexample of work performed by university faculty members to suggest thatwhen work behaviors are inherently nonprogrammable, an organization isforced to monitor behavior by assessing outcomes (e.g., number of publica-tions). Because a principal is actually buying an agent's behavior, it is mostefficient to reward the agent for actual behaviors when those behaviors canbe efficiently observed and evaluated.

Hypothesis 1: Managers in highly programmable jobs willreceive a lower proportion of their compensation in theform of variable pay (bonus) than managers in less pro-grammable jobs.

Organizational Turbulence

One factor that may affect the level of task programmability in an organ-ization is the level of organizational turbulence. Cameron, Kim, and Whetten(1987) defined organizational turbulence as nontrivial, rapid, and discontinu-ous change in an organization, brought about by events such as restructurings,downsizings, sales and spin-offs of assets, and acquisitions. Because thediscontinuity caused by the turbulence disrupts the knowledge of means-ends relationships that exists during more stable periods, it is more difficultfor an organization that is rapidly changing and undergoing high levels ofuncertainty to specify the behaviors that it needs its managers to perform.Classical organization theory would predict that turbulence will force evenorganizations in industries in which tasks are highly programmable (e.g., theairline industry) to be more financially flexible and to use ESOPs, bonuses,and so forth (Thompson, 1967). As turbulence increases, a principal mustbecome increasingly flexible. One means of increasing fiexibility is to shiftfrom a fixed-salary-based compensation system to one with a higher propor-tion of variable pay.

Hypothesis 2: Managers in highly turbulent organizationswill receive a higher proportion of their compensation inthe form of variable pay (bonus) than managers in morestable organizations.

Hypothesis 2, drawn from classical organizational theory (Thompson,1967), conflicts in part with agency theory's risk premium hypothesis. Al-though organizational turbulence can be viewed by a principal as requiringincreased organizational flexibility, it can also be viewed by an agent as ahigh-risk employment situation. In this situation, the agency theory riskpremium hypothesis comes into play. This hypothesis predicts that to attractand retain managers in high-risk situations, the principal will be forced toshare the cost of risky employment by paying a premium to the agent (Amer-nic, 1984; Eisenhardt, 1989). The principal will have to increase the agent'soverall total level of compensation in order to protect the agent from risk.

Hypothesis 3: Managers in highly turbulent organizationswill receive higher levels of total cash compensation (sal-


ary and bonus) than managers in organizations with lowerlevels of turbulence.

Expected Length of an Agency Relationship

Another factor related to agency theory that may affect a manager'scompensation is the expected length of his or her relationship with an organ-ization. Agency theory suggests that in long-term relationships, the principalgains more information about the agent's behavior and therefore can moreeasily have compensation contracts that are behavior-based (fixed) ratherthan outcome-based (variable) (Eisenhardt, 1989).

Although the actual tenure of the agent is important, it is possible toextend this argument to the expected tenure of the agent, insofar as the futureexpectations of the agent and the principal have a role in shapingthe compensation arrangements between them. Thus, if both the agent andthe principal expect and experience a long-term relationship, agency the-ory predicts compensation contracts will be less outcome-based (variable)(Eisenhardt, 1989; Lambert, 1983).

When organizations expect relationships with managers to last, theytend to have behavior-based control systems (Sonnenfeld & Peiperl, 1988;Sonnenfeld, Peiperl, & Kotter, 1988). These organizations are interested insocializing their managers into their culture and so tend to emphasize theirmanagers' behaviors rather than the outcomes the managers are accountablefor. Organizations that do not expect a long-term relationship to develop,and hence do not expend much effort in socializing their managers, put lessemphasis on behaviors and more on actual outcomes and would be expectedto place greater emphasis on outcome-based or variable pay. Simply put,when organizations place greater emphasis on longer-term relationships,they face lower perceived risks, and therefore, have less need to rely onvariable pay (e.g., the "old" IBM). When long-term relationships decline,risks go up and the organization tends to gravitate toward risk sharing (e.g.,the "new" IBM).

One way to identify an organization's expectations regarding the lengthof its relationship with its managers is through analyzing its human resourcepolicies. Employment security, clear promotion ladders, and investments intraining and development are all signals to managers that the organizationexpects to maintain a long-term relationship with them. Because the princi-pals are better able to monitor and assess the behaviors of their agents inlong-term relationships, managers in organizations with human resourcepolicies that encourage such relationships should receive a smaller propor-tion of their compensation in the form of variable pay than do managers inorganizations without such policies (Eisenhardt, 1989; Lambert, 1983).

Hypothesis 4: Managers in organizations with human re-source policies that encourage long-term relationships willreceive a smaller proportion of their salary in the formof variable pay than managers in organizations withoutsuch policies.

1996 Stroh, Brett, Baumann, and Beilly 755

Control Variables

A number of other variables at both the organizational and individuallevels may also influence managers' compensation in important ways andshould be controlled for in any study seeking to test agency theory predictionsabout compensation (James, Mulaik, & Brett, 1982). At the organizationallevel, an organization's industry membership (Gerhart & Milkovich, 1990;Mahoney, 1979), size (Deckop, 1988), and performance (Jensen & Murphy,1990; Leonard, 1990) should be taken into account.

Human capital theory suggests there are also important individual char-acteristics (Becker, 1975) that may affect individual performance. These in-clude educational level (an index of general skills), organizational tenure(an index of firm-specific skills), job tenure (an index of firm-specific skills),and labor market experience (an index of general skills).

METHODS

The opportunity to conduct this study emerged as a result of our conduct-ing a larger research project on relocated managers in Fortune 500 corpora-tions. We selected eight industries (pharmaceutical/hospital supplies, com-munications, consumer products/food, professional and financial services,retailing, hotel, chemical, and manufacturing) and at least two target compa-nies to represent each industry. Industries were selected to ensure that therewould be between-industry variance on the level of organizational tvu-bu-lence. Information from the general business press was used to make this se-lection.

Sample

Sample of organizations. The sample of organizations included 20 For-tune 500 companies ranging in size from 4,000 to 275,000 employees. Themean number of employees was 65,000, and average annual revenues were$12.2 billion. Because data were missing on a measure of organizationalperformance, three organizations were excluded from the study.

Sample of managers. In 1989, we surveyed lower- to middle-level man-agers who had been transferred by their companies within the previous twoyears (1987 and 1988). Surveys were mailed to 50 randomly selected manag-ers from each organization. With one follow-up phone call, the response ratewas 67 percent [n = 670).

We were concerned that, relative to the population of managers in gen-eral, ovir sample might underrepresent singles, women, and managers indual-career households with children, as a result of discrimination in theallocation of transfer opportunities or self-selection in the acceptance ofopportunities to transfer. Therefore, after randomly selecting managers fi:omeach firm, we randomly selected as many as 150 other managers from eachcompany's list, and we sent a letter to these people asking them to participatein the study if they fit into one of our special categories. This process madeour sample more representative of the population of managers in generaland increased the sample by 359 managers, for a total of 1,029 participants.


Two years later, in 1991, we again sent surveys to the homes of our 1,029respondents. With one follow-up phone call, we received 720 responses, fora response rate of 70 percent. After dropping cases with missing data (n =204), respondents who indicated that they were not in the managerial ranks(n = 90), and managers who were no longer with their 1989 employer [n =117), we were left with a sample size of 309 (see Table 1).

MeasuresProportion of variable pay. In the 1991 survey, managers were asked

their base salary and individual performance bonus for 1991. We divided

TABLE 1Demographic Characteristics of Respondents"

Characteristic

GenderMaleFemale

Age in 1989Education

High schoolSome collegeCollege graduateMaster's degreeOther degree

Marital statusMarried, cohabitingUnmarried

Number of childrenRace

WhiteBlackOther

Functional areaSales/marketingEngineeringProductionResearchAccounting/financeComputers/systemsAdministration/HROther

Tenure with 1989 employerNumber of geographic movesNumber of employers1989 salary1989 job level

NonmanagementLower managementMiddle managementUpper management

1989 Respondents

7 8 %

2236.5 years

3.4%12.861.219.0

2.9

76.4%23.5

1.0

95.8%1.6

2.6

37.1%8.9

8.3

3.8

10.24.6

12.713.9

10.3 years3.9

2.4

$57,752

17.6%27.045.5

9.8

Study Sample

82%18

38 years

3.4%12.358.522.1

3.6

81.7%18.3

1.2

96.0%1.1

2.9

31.6%7.5

8.7

3.8

11.64.2

13.214.9

11.4 years4.1

2.5

$62,210

31.358.110.6

' For the 1989 respondents, n = 1,029; for the study sample, n = 309.


the amount of the bonus by their total yearly cash compensation (salary plusbonus) to obtain the proportion of each manager's annual compensationthat was variable. Although we asked respondents for their "individual"performance bonuses, we are not confident that they discriminated amongindividual, unit-level, and organizational bonuses. To help isolate the hy-pothesized effects on the proportion of individual performance bonuses, weincluded organizational performance as a control variable.

Total cash compensation. Total cash compensation was measured onthe 1991 survey with two open-ended questions, one requesting base salary,the other requesting bonus compensation. We added results for these twovariables together and logged them to decrease heteroscedasticity (Boyd,1994; Finkelstein & Hambrick, 1989).

Task programmability. Kanter (1977) and others have noted that ingeneral, managers' behaviors are more difficult to define and monitor thanthose of lower-level employees. Thus, the task programmability measure wasoperationally defined with three survey questions from Quinn and Staines(1977). Managers were asked to respond to the following statements usinga four-point scale ranging from "often true" to "never true": (1) "My responsi-bilities are clearly defined" (indicating high task programmability), (2) "Ican see the results of my work" (indicating the organization's ability to defineoutcomes), and (3) "I am given a lot of freedom to decide how to do mywork" (low organizational ability to define behaviors). Alpha reliability was.77. A high score means high task programmability.

Turbulence. Archival data, summarized in Predicast's F &• S Index, U.S.Annual Editions, were used to measure the changes the companies in oursample had experienced. All were large organizations whose activities aretabulated regularly by Predicasts. We counted instances of changes in thefollowing categories for each organization: reduction in force, sale or spin-off of assets or operations, leveraged buyout, acquisition by another company,merger, joint venture, and attempted takeover. The relevant entries in Predi-cast's for 1989, the year of the first survey, were tabulated and counted(Miles & Huberman, 1984).

We formed an index by first summing the number of incidents in eachcategory and then, because we had no hypotheses about the different catego-ries, summing across categories (Stroh, Brett, & Reilly, 1994). Multiple articlesin Predicast's about the same change were counted only once. A high scoreindicated higher organizational turbulence.

To ensure data reliability, we used a second, independent coder to spot-check the content coding and counting. Intercoder reliability was 83 percent.To ensure the representativeness of the level of organizational turbulence inour sample of organizations, we selected a second sample of 17 companies,matched as closely as possible to our original sample in terms of industrymembership, annual sales, and number of employees. The level of organiza-tional turbulence was coded in the same way for this sample. There was nosignificant difference in the levels of turbulence in our two matched samples


[t = 1.80, p > .10), leading us to conclude that the level of turbulence inour analysis sample was representative.

Length of the agency relationship. Differences in the expected length ofthe relationship between the agent and the principal were measured withsurvey questions in which managers were asked about their career experi-ences with their organizations. The variables used for this index were theorganizational loyalty scale (Patchen, 1965), the career loyalty scale (Reilly,Brett, & Stroh, 1993), and a series of questions related to human resourcepolicies. The responses to these questions were measured on a five-pointLikert scale ranging from "strongly agree" to "strongly disagree." The ques-tions included "This division shows real concern for its people," "In thisdivision, people are treated as more important than things," "Employees arecarefully trained and developed over time in this division," "This divisionhas extensive employee and management training and development pro-grams," and "The job security is good." Items were worded to draw attentionto divisions because many organizations in our sample were multidivisional."Division" questions asked about the company's practices, and "company"questions asked respondents about their relationships with their companies.The index also included the number of years a manager expected to workfor a company.

To measure the length of the relationship, we used the entire 1989 dataset (JV = 1,029 managers). First, an index was constructed for each managerparticipating in the 1989 survey. Since the items composing the index wereon different scales, items were transformed into Z-scores. The coefficientalpha for the index was .81. High (positive) values corresponded to a long-term relationship between the manager and the organization. Second, weconducted an analysis of variance on the entire sample of 1,029 managersto confirm that different organizations had different means for the expectedlength of the agency relationship (cf. James, 1982). This analysis showedthat managers' career patterns were more similar within than between organ-izations. The index was then aggregated to the organizational level. Third,managers were assigned their organization's average for the length of relation-ship index. An aggregation is justified, according to James (1982), when thevariable of interest is an organizational policy rather than a construct thatexists only at the level of individual perceptions. Our operational definitionpresumes that organizations' human resources strategies are reflected in thecareer experiences of their managers. By reporting on their experiences withtheir organizations, our respondents acted as key informants concerning theexpected length of the agency relationship. Because we asked about therespondents' actual experiences and not their organizations' espoused poli-cies, we measured this organization-level agency theory construct as "real-ized" human resource strategies (Mintzberg, 1985).

Control variables. Industry membership was denoted by dummy vari-ables corresponding to an organization's primary industry. We chose themanufacturing industry as the reference category and excluded it from regres-sion analyses in order to avoid a singular matrix. The other industry dummy

1996 Stroh, Brett, Baumann, ond Beilly 759

variables are therefore interpreted relative to the manufacturing industry(Kerlinger & Pedhazur, 1973).

Organizational size was measured as the natural logarithm of the numberof employees, vi hich was obtained from a human resources liaison in eachorganization.

Organizational performance was measured by an index composed of anorganization's return on equity and its return on assets for 1989 (obtainedfrom COMPUSTAT tapes), along with division managers' ratings of corporateperformance. The division managers rated corporate performance relative totheir industry on four dimensions from Khandwalla (1977): long-run profit-ability, growth rate of sales or revenues, financial strength, and public image.The response format was a five-point Likert scale ranging from "very good"to "very poor"; the coefficient alpha for this scale was .73. We transformedthe three components (ROE, ROA, and division managers' ratings) into Z-scores and then them averaged to form an index of organization performance.

The educational level of the respondents was measured on a six-pointscale ranging from "high school" to "M.D., J.D., Ph.D., or other professionaldegree." Organizational tenure, job tenure, and labor market experience weremeasured by open-ended questions. Management level was measured by afour-point scale ranging from "nonmanagement" to "upper management."

Analysis

We tested our hypotheses using hierarchical multiple regression analy-sis, entering the organizational and individual control variables in the firststep of the regression. In the second step, the industry controls were entered.In the third step, task programmability, organizational turbulence, and ex-pected length of the agency relationship were entered.

RESULTS

Table 2 presents results for the analysis of the proportion of variablepay managers received (variable means and standard deviations are availableupon request). Model 1 regressed the proportion of variable pay on theorganization- and individual-level control variables. This model was signifi-cant (F730] = 2.59, p < .01, adjusted ff = .04). Next, in model 2, the industry-level controls were entered. The overall model was significant (F,4,294 = 7.57,p < .01, adjusted R^ = .23), as was the change in ff (Aff = .21, AF = 11.83,p < .01). Finally, in model 3, the study variables were entered. The modelwas significant (F,7,29i = 7.76, p < 01, adjusted ff = .27). The addition ofthe study variables resulted in another significant improvement in R''(Aff = .04, AF = 6.62, p < .01).

Table 3 presents results for the analysis of the risk premium hypothesis(total cash compensation). Model 1 regressed logged total cash compensationon the organization- and individual-level control variables. This model wassignificant (Fyjoi = 3.74, p < .01, adjusted ff = .06). Next, in model 2,we entered the industry-level controls. The overall model was significant

= 3.83, p < .01, adjusted ff = .11). The change in R^ was also significant


TABLE 2Results of Regression Analysis for 1991 Variable Pay as a Proportion of

Total Compensation"

Variables

Organizational performanceOrganizational sizeExperienceOrganizational tenureJob tenureEducationLevelChemicalFinancialPharmaceuticalRetailHotelCommunicationsConsumer productsTask programmability''Turbulence'^Expected length of agency relationship''Multiple B

Adjusted B'Change in ffFdf

Model 1 p

.00

.00-.00-.00

.01

.00

.02**

.24

.06

.04

2.59**7,301

Model 2 p

.01

.00

.00-.00

.00

.00

.02**-.01

.22**

.06**

.05*

.03-.02

.10**

.51

.26

.23

.21**7.57**

14,294

Model 3 p

.04**-.00

.00

.00

.00

.00

.02**

.03

.25**

.10**

.18**

.07**-.02

.09**- .01*

.02**-.05**

.56

.31

.27

.04**7.76**

17,291

" N = 309. The coefficients reported show differences in variable pay as a proportion oftotal compensation for a one-standard deviation change in continuous predictor variables. Thecoefficients for the industry dummy variables report differences in variable pay as a proportionof total compensation between the focal industry and manufacturing.

'' A high value means higher task programmability." A high value means higher levels of turbulence.'' A high value means longer organizational tenure.

* p < .05** p < .01

(Aff = .07, AF = 3.69, p < .01). Finally, in model 3, the study variableswere entered. The addition of the study variables did not result in a significantimprovement in ff (Aff = .01, AF = 1.35, p < .26).

The results shown in Table 2 support Hypothesis 1: In organizationswith high levels of task programmability, there was less use of variable payO = - . 0 1 , p < .05). A one-standard-deviation increase in the level of taskprogrammability was associated with a one percentage point decrease in theproportion of variable pay. As predicted by agency theory, organizationsappear to be using variable pay as an incentive to align the goals of theirmanagers to those of the organizations when tasks are less programmable.

Hypothesis 2 was also supported. Managers in organizations with highlevels of turbulence received a higher proportion of their compensation in


TABLE 3Results of Regression Analysis for 1991 Total Cash Compensation"

Variables

Organizational performanceOrganizational sizeExperienceOrganizational tenureJob tenureEducationLevelChemicalFinancialPharmaceuticalRetailHotelCommunicationsConsumer productsTask programmability''Turbulence'^Expected length of agency relationship''Multiple B

Adjusted B^Change in B^F

df

Model 1 p

.03-.13**

.03

.08-.04

.06

.14**

.28

.08

.06

3.74**7,301

Model 2 p

-.12-.09

.04

.06

.01

.03

.17**-.01

-1.25**.25

-.32-.10-.13

.06

.39

.15

.11

.073.83**

14,294

Model 3 p

-.17-.01

.04

.05

.05

.04

.17**-.08

-1.24**.15

-.64-.30-.13

.04-.04-.10

.07

.41

.16

.12

.013.40**

17,291

" The coefficients reported show differences in the logarithm of total compensation for aone-standard deviation change in continuous predictor variables. The coefficients for theindustry dummy variables report the differences in total cash compensation between the focalindustry and the manufacturing industry. Af = 309.

'' A high value means higher task programmability.'= A high value means higher levels of turbulence.'' A high value means longer organizational tenure.

* p < .05**p < .01

the form of variable pay [ft = .02, p < .01). A one-standard-deviation increasein the level of turbulence was associated with a two percentage point increasein the proportion of variable pay. That is, in organizations with high levelsof uncertainty, the compensation strategies tended to reflect this financialuncertainty. The data support the notion that organizations in turbulentenvironments have a greater need for more flexible (less programmable)compensation strategies. Therefore, organizations with high levels of uncer-tainty did show a greater reliance on the use of variable pay to ensure appro-priate behaviors in their managers. This finding is consistent with the predic-tions of classical organization theory (Thompson, 1967).

The competing risk premium hypothesis (Hypothesis 3) was not sup-ported (see Table 3). Total cash compensation strategies did not significantlydifferentiate managers in turbulent organizations from managers in less tur-


bulent organizations ()3 = -.10, p < .09). Consequently, in organizationswith high levels of turbulence, the firms were shifting the risk of the turbulentenvironment to the managers through increased variable pay, instead ofsharing the cost of the risk by giving higher total compensation, as waspredicted by agency theory's risk premium hypothesis.

Hypothesis 4 was supported (Table 2). An expectation of a long-termagency relationship was negatively associated with the use of variable pay(/3 = - .05, p < .01). A one-standard-deviation increase in the expectationof a long-term agency relationship was associated with a five percentagepoint decrease in the proportion of variable pay. That is (regardless of thesize of the organization, which was controlled for), those organizations whosehuman resource policies and practices supported the development of long-term relationships had compensation strategies that reflected their ability tobetter monitor and assess behavior, and they were therefore less likely touse variable pay as an incentive to achieve certain behaviors.

The effects of the control variables were primarily consistent with priortheory and research. Higher organizational performance was associated witha higher proportion of variable pay. Management level was more importantthan tenure with a firm in predicting proportion of variable pay: highermanagement level was associated with a higher proportion of variable pay.Relative to the manufacturing industry (which was the reference category),the financial, pharmaceutical, retail, hotel, and consumer products industriesused a higher proportion of variable pay. Organizational size was not relatedto the use of variable pay.

DISCUSSION

The results of this study support agency theory predictions that proximalorganizational level constructs affect compensation strategies. The level ofa manager's task programmability was negatively associated with the use ofvariable pay: when tasks were less programmable, there was a greater relianceon variable pay compensation. The results did not, however, support agencytheory predictions regarding the risk premium hypothesis. When organiza-tions were particularly turbulent, characterized by high levels of risk anduncertainty, they did not buffer their managers' risk by paying them a riskpremium (a higher total cash compensation). Rather, our data showed thatorganizations in which turbulence was greater shifted the financial risk totheir managers by paying proportionately higher levels of variable pay. Re-gardless of whether tasks are programmable or not, in turbulent environmentsorganizations choose to use variable compensation strategies apparently tobetter deal with their own turbulence-imposed risks. Organizations shift therisk to the managers in the form of variable pay, as predicted by classicalorganization theory.

The expectation of a long-term relationship between an organizationand its managers was negatively associated with the use of variable pay. Ourdata support agency theory predictions that when such an expectation exists


there is a greater reliance on behavior-based compensation, perhaps becausemore information about an agent's behavior is available.

Many of the control variables were also associated with variable pay inthe expected directions. Thus, high organizational performance was associ-ated with the use of higher proportions of variable pay. This finding couldmean that using a variable pay compensation strategy may pay off in termsof an organization's performance. Alternatively, better performers may havemore slack resources to distribute to managers as variable pay.

Organizations in the manufacturing industry gave their managers a lowerproportion of variable pay than organizations in many other industries. Thisfinding is consistent with the assumption that large manufacturing organiza-tions may be more capital-intensive than other companies and the findingthat the capital intensity of an organization's technology is generally in-versely related to the use of variable pay (cf. Gomez-Mejia & Balkin, 1992a).

The effects of the human capital variables (labor market experience,organizational tenure, job tenure, and education) were notably inconsistentwith prior theory and research. After controlling for other factors, we didnot find that these variables were associated with the proportion of variablepay that managers received. This result, too, is reasonable if we suppose thatthe main role of human capital variables is to sort people into positions, ata certain hierarchical level, in a certain organization, in a certain industry,and so on, and that these position-related factors have more direct effectson the proportion of variable pay that managers receive.

Limitations of the Study

Like all theory-testing studies, this study measured key theoretical con-structs indirectly. Behavior-based measures would be useful to validate theresults of this study. Also, the sample for the estimate of human resourcepolicies consisted of managers who had recently been relocated by theirorganizations. Although these managers may have actually provided an ex-cellent view of an organization's human resource policies because they werepeople who had made a commitment to the organization by accepting atransfer, their use as respondents may also limit the generalizability of the re-sults.

In our study, compensation data for only one year were used to measurethe proportion of variable pay managers received. A more reliable measure-ment would be possible if data for multiple years were combined. In addition,although we asked respondents to report their "individual performance bo-nuses," they may not have been able to discriminate between individual-,unit-, and organization-level bonuses. However, our use of organizationalperformance as a control variable helped isolate the effects of the studyvariables on the individual component of the performance bonus.

Another potential limitation of the study is that variables like organiza-tional size and performance, found to be correlated with top managers' payin other studies (Deckop, 1988; Jensen & Murphy, 1990; Leonard, 1990), werenot correlated with pay in this study. It may be that organizational size and


performance are too distal to have a significant impact on middle managers'pay. Alternatively, our sample organizations may not have been large enoughto capture variance in these macro variables that is related to middle manag-ers' pay. We have tried to phrase our hypotheses and findings in a mannerconsistent with an acknowledgment of this limitation.

Lastly, the sample of managers in the study was not drawn to be represen-tative of industries. In some categories, such as financial services, our less-well-paid sample does not reflect the data from the industry as a whole. Thislimitation results from our sampling managers who had been transferred. Inthe financial services industry, these tended to be corporate managers, notinvestment bankers.

Despite these cautions, the results contribute to the development ofmethods in compensation research and to understanding of agency theory'spower in accounting for middle manager compensation.

Contributions of the Study

Methodological contributions. We define variable pay slightly differ-ently than did Gerhart and Milkovich (1990). Where they defined "pay mix"as the ratio of bonus to salary, we have used the proportion of variable payto total compensation. We believe that our measurement more accuratelycaptures the construct of interest, which is the level of variability in thecompensation package. The properties of proportions are also more attractivethan those of ratios: proportions are bounded between zero and one andso are readily interpretable, but ratios range from zero to infinity (i.e., the.proportion of bonus to salary for a manager who received a very small salaryrelative to bonus would be near one but the ratio would be approachinginfinity). Therefore, our coefficients have a straightforward interpretation.

A second methodological strength of the study is the lag in time betweenthe measurement of the agency theory constructs and the measurement ofmanagers' compensation. It is likely that there will be some lag before inde-pendent variables are reflected in a dependent variable (James et al., 1982).For instance, a manager's compensation in a current year may reflect theprevious year's performance, simply because year-end results must be ob-tained before complete performance data can be compiled. Furthermore,there is likely to be some rigidity in a pay system (Doeringer & Piore, 1971;Leonard, 1990), which makes an instantaneous response to changing condi-tions unlikely.

A third contribution of this study is the testing of agency theory hypothe-ses on a sample of middle managers.

Theoretical contributions. The study supports agency theory's predic-tions relating middle managers' task programmability and the expectedlength of their agency relationships to variable pay. The study does notconfirm agency theory's risk premium hypothesis for middle managers. In-stead, the risk shifting hypothesis of classical organization theory was con-firmed. Middle managers whose organizations were turbulent received agreater proportion of variable pay, but no greater overall compensation, than

1996 Stroh, Brett, Baumann, and Reilly ' 765

middle managers whose organizations were less turbulent. These resultssuggest, counter to agency theory, that middle managers do not receive arisk premium when an organization's conditions are turbulent but that in-stead, the organization shifts that risk to the managers in the form of variablepay. The findings are consistent with work of Balkin and Gomez-Mejia (1987)showing that organizations in turbulent environments have greater risk shar-ing (are more likely to use variable pay) than those in stable environmentsand do not compensate for the managers' risk in the form of higher base pay.

Managerial and policy implications. Findings from this study provideimportant information for those developing or implementing managerial pol-icy. Both the popular press and empirical research are asking how managers'commitment to their organizations can be increased, given the high levelsof turbulence in the corporate environment. Findings from our hypothesison the expected length of an agency relationship show that managers whosecompensation is less tied to variable pay do expect to stay with their organiza-tions longer. Consequently, organizations that prefer to buy their labor ontheir internal labor market and also prefer long-term managerial relationshipsand commitment should implement compensation strategies that rely lesson variable pay. The reverse strategy could also be implemented. Organiza-tions that prefer high levels of turnover (to "bring in new blood," perhaps)should implement compensation strategies that have a higher variable paycomponent.

This same logic would apply to organizations wanting to retain talentedmanagers during times of high levels of turbulence. Findings from this studysuggest that middle-level managers do not receive a risk premium duringhighly tubulent periods. Implementing a compensation package that empha-sizes the managers' total compensation package (rather than variable pay)should help to retain talented managers that the organization wishes to retainduring these turbulent times.

In conclusion, this study supports, extends, and refutes agency theory.Key agency theory constructs (task programmability, risk premium, thelength of the agency relationship) were measured as job characteristics, busi-ness strategy (organizational turbulence), and realized human resource strat-egy. These constructs were related to an important facet of an organization'scompensation strategy, the use of variable pay compensation. However, un-like research on upper-level executives, this research did not support agencytheory's risk premium hypothesis.

REFERENCES

Alchian, A. A., & Demsetz, H. 1972. Production, information cost, and economic organization.American Economic Review, 62: 777-795.

Amernic, J. H. 1984. A look at agency theory for the novice: Part I. CAmagazine, November:90-97.

Baker, G. P., Jensen, M. C, & Murphy, K. J. 1988. Compensation and incentives: Practice vs.theory. Journal of Finance, 43: 593-616.


Balkin, D. B., & Gomez-Mejia, L. R. 1987. Toward a contingency theory of compensation strategy.Strategic Management Joumal, 8: 169-182.

Balkin, D. B., & Gomez-Mejia, L. R. 1990. Matching compensation and organizational strategies.Strategic Management Joumal, 11: 153-169.

Beatty, R. P., & Zajac, E. 1994. Managerial incentives, monitoring, and risk bearing: A studyof executive compensation, ownership, and board structure in initial public offerings.Administrative Science Quarterly, 39: 313-335.

Becker, G. 1975. Human capital: A theoretical and empirical analysis, with special referenceto education (2nd ed.). Chicago: University of Ghicago Press.

Boyd, G. K. 1994. Board control and CEO compensation. Strategic Management Journal,15: 335-344.

Gameron, K., Kim, M. Y., & Whetton, D. A. 1987. Organizational effects of decline and turbulence.Administrative Science Quarterly, 32: 222-240.

Deckop, J. R. 1988. Determinants of chief executive officer compensation. Industrial and LahorRelations Review, 41: 215-226.

Doeringer, P. B.,& Piore, M. J. 1971. Internal labor markets and manpower analysis. Lexington,MA: Heath Lexington Books.

Eisenhardt, K. M. 1988. Agency- and institutional-theory explanations: The case of retail sales.Academy of Management Joumal, 31: 488-511.

Eisenhardt, K. M. 1989. Agency theory: An assessment and review. Academy of ManagementReview, 14: 57-74.

Finkelstein, S., & Hambrick, D. C. 1989. Chief executive compensations: A study of the intersec-tion of markets and political processes. Strategic Management Journal, 10: 121-134.

Gerhart, B., & Milkovich, G. T. 1990. Organizational differences in managerial compensationand financial performance. Academy of Management Joumal, 33: 663-691.

Gomez-Mejia, L. R., & Balkin, D. B. 1992a. Compensation, organizational strategy, and firmperformance. Gincinnati: South-Western.

Gomez-Mejia, L. R., & Balkin, D. B. 1992b. Determinants of faculty pay: An agency theoryperspective. Academy of Management Joumal, 35: 921-955.

Gomez-Mejia, L. R., & Welbourne, T. M. 1988. Compensation strategy: An overview and futuresteps. Human Resource Planning, 11: 173-189.

James, L. R. 1982. Aggregation bias in estimates of perceptual agreement. Jounal of AppliedPsychology, 67: 219-229.

James, L. R., Mulaik, S. A., & Brett, J. M. 1982. Causal analysis: Assumptions, models, anddata. Beverly Hills, CA: Sage.

Jensen, M. C, & Murphy, K. J. 1990. Performance pay and top-management incentives. Journalof Political Economy, 98: 225-264.

Kanter, R. M. 1977. Men and women of the corporation. New York: Basic Books.

Kerlinger, F. N., & Pedhazur, E. J. 1973. Multiple regression in behavioral research. New York:Holt, Rinehart & Winston.

Khandwalla, P. N. 1977. The design of organizations. New York: Harcourt, Brace, Jovanovich.

Lambert, R. A. 1983. Long-term contracts and moral hazard. Joumal of Economics, 14: 441-452.

Lambert, R. A., Larcker, D. F., & Weigelt, K. 1993. The structure of organization incentives.Administrative Science Quarterly, 38: 438-461.

Lawler, E. E., IIL 1981. Pay and organization development. Reading, MA: Addison-Wesley.


Lawler, E. E., III. 1990. Strategic pay: Aligning organizational strategies and pay systems.San Francisco: Jossey-Bass.

Leonard, J. S. 1990. Executive pay and firm performance. Industrial and Labor RelationsReview, 43: 13-29.

Mahoney, T. A. 1979. Compensation and reward perspectives. Homewood, IL: Irwin.

Mellow, W. 1982. Employer size and wages, flevien' of Economics & Statistics, 64: 495-501.

Miles, M. B., & Huberman, A. M. 1984. Qualitative data analysis: A sourcebook of methods.Beverly Hills, CA: Sage.

Mintzberg, H. 1985. Strategy and its formation. Working paper, McGill University, Montreal.

Patchen, M. 1965. Some questionnaire measures of employee morale: A report on theirreliability and validity. Monograph no. 41, Institute for Social Research, Ann Arbor, MI.

Quinn, R. P., & Staines, G. L. 1977. The 1977 quality of life employment survey. Ann Arbor,MI: Institute for Social Research.

Reilly, A. H., Brett, J. M., & Stroh, L. K. 1993. The impact of corporate turbulence on managers'attitudes. Strategic Management Joumal, 14: 167-179.

Sonnenfeld, J. A., & Peiperl, M. A. 1988. Staffing policy as a strategic response: A typology ofcareer systems. Academy of Management Review, 13: 588-600.

Sonnenfeld, J. A., Peiperl, M. A., & Kotter, J. P. 1988. Strategic determinants of managerial labormarkets: A career systems view. Human Resource Management, 27: 369-388.

Stroh, L. K., Brett, J. M., & Reilly, A. H. 1994. A decade of change: Managers' attachment totheir organizations and their jobs. Human Resource Management Journal, 33: 531-548.

Thompson, J. D. 1967. Organizations in action. New York: McGraw-Hill.

Westphal, J. D., & Zajac, E. J. 1994. Substance and symbolism in GEOs' long-term incentiveplans. Administrative Science Quarterly, 39: 367-390.

Linda K. Stroh (Ph.D., Northwestern University) is an associate professor of organizationbehavior at the Institute of Human Resources and Industrial Relations, Loyola UniversityGhicago. Her work focuses primarily on organizational policies impacting managers'careers, attitudes, and commitment to the organization, both domestically and interna-tionally.

Jeanne M. Brett is the DeWitt W. Buchanan, Jr., Professor of Dispute Resolution andOrganizations, at the Kellogg Graduate School of Management, Northwestern University.She holds a Ph.D. degree in psychology from the University of Illinois. She studiesmanagerial career development, particularly as it relates to issues of work and family.

Joseph P. Baumann is a graduate student in organization behavior at the Kellogg Gradu-ate School of Management, Northwestern University. He holds a B.A. degree from theUniversity of Delaware, where he was a psychology major. His current research is inthe area of compensation.

Anne H. Reilly is an associate professor in the Department of Management at LoyolaUniversity Ghicago. She earned her Ph.D. degree in organizational behavior from North-western University. Her current research interests include organizational change andwork and family issues.

Agency Theory and Variable Pay Compensation Strategies.

Documents

Transcript of Agency Theory and Variable Pay Compensation Strategies.