Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.
Management Information Systems Research Center, University of Minnesota is collaborating with JSTOR to digitize,
preserve and extend access to MIS Quarterly.
http://www.jstor.org
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
Factors
Affecting
Developers'
Software
Performance:
Integrated
An
Approach'
Backgroundand Introduction
Abstract
Software developers' performancehas a direct
An
impacton softwaredevelopmentproductivity.
understandingof the factors thataffect this performancecouldhelp determinewhereto concentrate management efforts (and related financial
resources)froma practicalstandpoint,and where
to focus research efforts from an academic
perspective.Togain furtherinsightintothese factors, this study extends prior research by integrating elements from expectancy theory,
goal-settingtheory,and organizationalbehavior
specific to the software development process.
Theresearchresultsprovidenew insightsregarding the relative importanceof how expectancy
theory, goal-setting theory, and individual
characteristicsaffect the perceivedperformance
of software development professionals. These
preliminaryfindings indicate that goal-setting
theory may have complex implicationsfor software developmentperformance. Goal difficulty
has a negative relationshipto performancebut
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
Conceptual Framework
Integrated research model
The research just cited shows the effects of
several independentvariables on performancerelated dependent variables using separate
behavioralmodels. Ourresearchdevelops an integrated model that permitsthe assessment of
Expectancy theory
Expectancytheory continues to be widely used
to examine motivationalissues (Baker, et al.,
1989; Brownelland Mclnnes, 1986; Butlerand
Womer, 1985; Harrelland Stahl, 1984; Kaplan,
1985; Nickerson and McClelland,1989). Much
of the interest in expectancy theory research
stems from the belief that highly motivated individualswill exert higher effortlevels and consequently will tend to performat higher levels
thantheirless motivatedcontemporaries.The expectancy theory component of the research
model is shown in Figure la.
As shown in the equation, performance(P) can
be described as a function of an individual's
effort-level(E), ability(A), and role perceptions
(R) (Ferris, 1977; Lawlerand Suttle, 1973).
P = f (E, A, R)
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
Goal-settingtheory
Goal-settingtheory research shows that expectancies about the level of success may also be
related to the clarity and difficultyof work requirements(Carrolland Tosi, 1973; Locke and
Latham, 1990; Locke, et al., 1981). The goalsetting theorycomponentof the research model
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
Path 8
398
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
Individualcharacteristics
Softwaredevelopers' performancemay also be
affected by individualcharacteristics. The individualcharacteristicsdimensionof the research
model is shown in Figure lc. Performancehas
been relatedto an individual'sneed for achievement, locus of control,and self-esteem (Mitchell,
1974; 1979; 1982). Need for achievement is the
extent to which the person values success
(McClelland,1961). Individualswithhighachievement needs preferto workin situationswherethe
desired resultsare clear. Individualachievement
needs have been shown to be positivelyrelated
to both effort (Path 2a) and performance(Path
2b) (Mitchell, 1979; Porter and Lawler, 1968;
Steers, 1975; Steers and Porter,1983). Locus of
control refers to whether people believe that
their fate is controlledby external factors or by
the people themselves. There is evidence that
people with a strong internal locus of control
(where they believe they controltheir own fate)
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
exert greater effort(Path 3a) and higher performance (Path3b) than those witha strong external locus of control(wherethey believe theirfate
is controlledby others) (Anderson, 1977). Selfesteem is a person's sense of self-worth.High
self-esteem is proposed to be positivelyrelated
to both effort (Path 1b) and performance(Path
la). Evidence shows that managers with high
self-esteem reporthigher levels of performance
when they have difficultgoals to achieve than
those with low self-esteem (Carrolland Tosi,
1973). However, argument shows that an individual'sneed forachievementand self-esteem
may moderatethe relationshipbetweenworkperformanceand the characteristicsof workgoals.
Method
Subjects
The data for this study were gathered at three
major software development organizations
locatedin the midwest,southwest,and southeast
United States. These firms are involvedwith a
wide spectrumof softwaredevelopmentprojects
ranging from embedded systems on classified
governmentcontractsto telecommunicationsfor
publictelephonesystems and payrollsystems for
business organizations.All three firms are advocates of the principlesof softwareengineering,
and they use the term software engineer interchangeably with software developer. The
respondentsactuallyparticipatedinthe totalsoftware developmentprocess fromproblemdefinition through debugging and implementation.
Because these three firmssupportthe concept
of theirsoftwaredevelopers' involvementwitha
project from its beginning to its end, the
respondents provideda homogeneous sample
knowledgeable of the entire software development process. The evidence of this homogeneity was apparent from the separate analyses
performed for each company. There were no
statisticallysignificantdifferences between the
demographics,individualcharacteristics,or other
variables measured for the respondents.
Responses were thereforecombinedacross the
three firmsforthe subsequent analyses reported
below.Thisfindingalso providedevidence forthe
generalizationof the results across at least the
three firm environments. These responses,
however, might differfrom those of subjects in
Measures
A preliminaryquestionnairewas pilot-testedwith
17 softwaredevelopers to assess logical inconsistencies, ease of understanding, and task
relevance.Therewere some modificationsto the
originalinstrumentto clarifythe meaning of particularsections. None of these responses were
used in the analysis reported in this study.
Performance
Perceived performance was measured by a
single item, which asked each subject to com-
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
Percent
Gender
Female
Male
89
246
27
73
Education
HS Diploma + Some College
Bachelors Degree
BS + Some GraduateWork
Masters Degree
PhD
53
153
67
60
2
16
45
20
18
1
Category
Age
Mean
Standard Deviation
Software Engineering Experience
Mean
Standard Deviation
36.4 years
8.2 years
9.8 years
6.9 years
of self-appraisal-basedperformanceevaluation
concludes that lenient self-appraisalsare likely
to occur when they lead to some personal gain
or when no independentperformancemeasures
are available (Farh, et al., 1988). The level of
ratings, however, should not influence validity
unless there is severe restriction-of-range.
Anotherinvestigationon assessing the utilityof
self-evaluationsof work performancesuggests
that in theoreticalstudies of job performance,the
researcher is probablywell-advised to include
self-descriptionsas a source of data (Thornton,
1980).
A characteristicof the performancemeasure is
that it is a single item, ratherthan a multiple-item
measure. The use of a single-item measure for
performancecontinuesto be a topic of academic
controversy in the psychological literature
(Cascio, 1987). This was not foreseen as a problem forthis study, however,because it has been
demonstratedthat single items that capture the
global meaningof a concept may be effective indicators of performance(Scarpello and Campbell, 1983). While it is not possible to estimate
reliabilitywith conventional methods such as
alpha coefficients (Cronbach, 1951), there is
some evidence of the validityof this measure for
this study.Itwas possibleto correlatethe respon-
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
401
PredictingDevelopers' Performance
Ability
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
Individual Characteristics
Three individualcharacteristicswere measured:
need forachievement, locus of control,and selfesteem. Need forachievementwas measuredusing a five-itemscale developed by Steers and
Braunstein (1976) (a=0.73). Locus of control
refersto the degree to whichindividualsfeel they
can controltheirown "destiny."Individualswith
a stronglocus of controlfeel thatmost thingsthat
happen to them are influenced by themselves,
whereas individualswith a low locus of control
feel they have littleor no controlover things that
happento them. Rotter's(1966) 15-itemmeasure
of locus of control was used for this study
(a =0.71). Self-esteem was measured with the
25-five item adult formof Coopersmith's(1967)
Role Ambiguity
Role ambiguitywas measured using the six-item
role ambiguityscale developed by Rizzo, et al.
(1970). Role ambiguityis the degree to whichthe
individualhas knowledgeaboutwhat behavioris
expected and whetherthere are guidelines concerning appropriate behavior. The internal
reliability (Cronbach, 1951) was very high
Analysis
The simple correlationmatrixforthe variablesis
shown in Table 2. Structuralequation modeling
using LISRELVII(Joreskog and Sorbom, 1988)
was appliedto test the fit of the research model
(shown in Figure 1) to the data collected using
sample covariance matrices. Two-stage least
squares and maximum-likelihoodprocedures
yielded initial and final parameter estimates,
respectively. This analysis provides direct
measures of the degree to whichtheoreticalconstructs are related, the extent of errors in
variables and equations, and the relationships
of
between constructsand the operationalization
(a = 0.89).
Goal Attributes
Thereis ampletheoryand evidence thatspecific,
challenging goals lead to higher performance
than general, easy goals (Carrolland Tosi, 1973;
Lathamand Locke, 1979; Locke, et al., 1981).
The Task-Goal AttributesScale developed by
Steers (1975)was used to measuregoal specificity (a =0.83) and goal difficulty(a =0.81).
PE
Variable
EF
RA
AN
Performance
Effort
(PE) 1
(EF) .38*
Role Ambiguity
AchievementNeeds
Locus of Control
1
(RA) -.15* -.32*
1
.39* -.25*
(AN) .40*
.20*
-.08
-.11*
-.03
(LC)
Self-Esteem
(SE)
Goal Specificity
Goal Difficulty
(GS) .08
(GD) -.01
.23*
Intellectual
(IA) .61*
Ability
Qualityof Education (QE) .17*
LC
SE
GS
GD
IA
QE
1
.20*
.15*
-.23*
.26*
1
-.30*
.20* -.79*
.23* -.03
.22*
1
.15* -.19*
.13*
.16* -.11* - .01
.19* .01
.11* -.06
.27* -.03
.14* .06
.05
.07
-.06
.01
.01
.07
*Statistically
significantat p < 0.05.
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
403
PredictingDevelopers' Performance
Results
Resultsof the analysisare shown as the empirical
performance model (Figure 2). The proposed
linkages of the research model (Figure 1) were
tested at a statisticalsignificancelevelof p <0.05.
Allsignificantlinkages were retainedin the empiricalperformancemodel.
The strength of the relationships among the
variables in the empirical performance model
(Figure2) is represented by standardizedsolution coefficients. These coefficientscould range
from -1 to +1 and provide meaningful
measuresto assess the relativestrengthbetween
variables.The empiricalperformancemodel explains 54 percent of the variance in the reported
individualperformancelevels (R2= .54).
The total effect of a variable on software
developers' perceptions of performanceis due
to both direct and indirecteffects of intervening
variables.An individual'sachievementneeds, for
example, have a totaleffect of 0.26 on perceived
performance(directeffect of 0.18 and indirecteffect of 0.08). Indirecteffects are calculated by
combiningtheireffects withinterveningvariables.
The indirecteffect of achievementneeds on performance is computed by multiplyingthe direct
effect of achievement needs on effort by the
direct effect of efforton performance.The total
effects on performance for all variables are
shown in Table 3.
Discussion
Priorto discussing the results, there are three
methodologicalissues in this study that warrant
mention.First,this researchemploysa structural
equation modeling technique, but because the
data are cross-sectional,the causal linksbetween
the variablesshould be considered withcaution
and subjectedto additionalfutureresearch.Second, with scales of the type used here there is
the possibility of common method variance,
whichwouldinflatethe relationshipsbetweenthe
variables.This does not appear to be a problem
with this data for two reasons. First, if common
methodvariancewas a serious problem,the zero
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
Relation
Achievement Needs
With Performance
Direct
Effect
Indirect
Effect
Total
Effect
0.18
0.08
0.26
0.15
----
0.15
0.11
----
0.11
With Performance
----
0.04
0.04
Goal Difficulty
With Performance
- 0.11
0.04
-0.07
Self-Esteem
With Performance
Locus of Control
With Performance
Goal Clarity
Effort
With Performance
0.21
---
0.21
0.54
----
0.54
Ability
With Performance
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
References
Anderson, C. "Locus of Control, Coping
Behaviorsand Performancein a Stress Setting," Journal of Applied Psychology (62:1),
February1977, pp. 446-451.
Baker, D., Ravichandran,R., and Randall, D.
"ExploringContrastingFormulationsof Expectancy Theory,"Decision Sciences (20:1),
Winter1989, pp.1-13.
Baroudi,J. "The Impactof Role Variableson IS
Personnel Work Attitudes and Intentions,"
MIS Quarterly(9:4), December 1985, pp.
341-356.
Boehm, B. "ImprovingSoftware Productivity,"
Computer(20:9),September 1987, pp. 43-57.
Brownell,P. and Mclnnes, M. "BudgetaryParticipation, Motivation, and Managerial
Performance,"TheAccountingReview(62:4),
October 1986, pp. 587-600.
Butler,J. and Womer,K. "Hierarchicalvs. NonNested Tests for Contrasting ExpectancyValence Models: Some Effects of Cognitive
Characteristics," Multivariate Behavioral
Research (20:3), July 1985, pp. 335-352.
Carroll,S. and Tosi, H. Managementby Objectives, McMillan,New York, NY, 1973.
Cascio, W. Applied Psychology in Personnel
Management,PrenticeHall,EnglewoodCliffs,
NJ, 1987.
Coopersmith, S. The Antecedents of SelfEsteem, Freeman,San Francisco,CA, 1967.
Couger, J. "Motivation Norms for Software
Engineers Versus Those for Programmer
Analysts," Proceedings of the 7th International Conference on InformationSystems,
December 15-17, 1986, San Diego, CA, pp.
214-223.
Couger, J. and Zawacki, R. Motivatingand
Managing ComputerPersonnel, John Wiley
& Sons, New York, NY, 1980.
Cronbach, L. "CoefficientAlpha and the Internal Structureof Tests," Psychometrika(16:3),
September 1951, pp. 297-334.
Curtis,B. "By the Way, Did Anyone Study Any
Real Programmers?"in EmpiricalStudies of
Programmers, E. Soloway and S. lyengar
(eds.), Alex PublishingCorp., Norwood,NJ,
1986, pp. 256-262.
Curtis, B., Krasner,H., and Iscoe, N. "A Field
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
of theACM
LargeSystems," Communications
(31:11), November 1988, pp. 1268-1287.
Farh, J. and Werbel, J. "The Effects of Purpose
of the Appraisal and Expectation of Validation
on Self-Appraisals Leniency," Journal of Applied Psychology (71:3), August 1986, pp.
527-529.
Farh, J., Werbel, J., and Bedeian, A. "An Empirical Investigation of Self-Appraisal-Based
Performance
Personnel
Evaluation,"
Psychology (41:1), Spring 1988, pp. 141-156.
Ferris, K. "A Test of the Expectancy Theory of
Motivation in an Accounting Environment,"
408
OrganizationalDynamics(8:2),Autumn1979,
pp. 68-80.
Lawler, E. and Suttle, J. "Expectancy Theory and
Job Behavior,"OrganizationalBehaviorand
Human Performance (9:3), June 1973, pp.
482-503.
Locke, E. and Latham, G. Goal Setting: A Motivational Technique That Works, Prentice Hall,
Englewood Cliffs, NJ, 1984.
Locke, E. and Latham, G. A Theory of Goal Setting and Task Performance, Prentice Hall,
Englewood Cliffs, NJ, 1990.
Locke, E., Saari, L., Shaw, K., and Latham, G.
"Goal Setting and Task Performance: 19691980," Psychological Bulletin (90:1), July
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
Appendix A
Motivation Level Information for Current Job
Your Current Job
In the followingsituation,you are asked to providelikelihood(probability)informationfor your current
job. Any response ranging from p = 0% (never) to p = 100% (always) is appropriate. If, however, you
If you are a highly effective performerin your currentjob, the likelihood(probability)that you will
-frequently work overtimeto meet deadlines is ...........................
(p =
(p =
(p =
(p =
.. (p =
Average
Effort
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
9
Great
Effort
PredictingDevelopers' Performance
Appendix B
Analysis Using the LISRELModel
In this paper, the Joreskog-Keesling-Wiley model, commonly known by the copyrighted name of the
computer program LISRELVII(Joreskog and Sorbom, 1988), was used. Detailed results of the analysis
r + rF + s
(A1)
Where:
=
effort
performance
self-esteem
achievement needs
locus of control
t =
goal difficulty
goal clarity
ability
.00
.21
.00
.00
.00
.15
.39
.18
.00
.11
.19
-.11
.19
.00
.00
.54
.71
.46
Where r represents a vector of latent dependent variables, t represents a vector of latent independent
variables, A and r are coefficient matrices, and s is a random vector of residuals (errors in equations,
randomdisturbanceterms).
Based on the proposed Integrated Research Model (Figure 1) 3(1,1), 0(1,2) and 0(2,2) were initially set
to zero to test the proposed relationship between effort and performance. Since there are no proposed
relationships between goal clarity and performance or effort and ability, r(2,5) and r(1,6), respectively,
were also set to zero. Following the intitialdata analysis, r(1,1) and r(1,3) were also set to zero because
there was no statistically significant relationship (p<.05) between effort and either self-esteem or locus
of control.
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
Ay defines the structural relationship between the latent dependent variables and their indicator
ment model:
y = Ayn + e
(A2)
Where:
effort level
perceived performance
Ay =
effort
performance
.00
e =
.00
Because the indicator variable for effort was captured using a behavioral decision-making scenario and
the indicator variable for performance was a single-item measure, it is not possible to estimate their
internal error terms, e. As is customary in LISREL analysis, these error terms are explicitly set to zero.
As a point of interest, it should be noted that this procedure is exactly the same as is done implicitly
self-esteem
need for achievement
locus of control
goal difficulty
goal clarity
intellectualability
qualityof education
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
PredictingDevelopers' Performance
100000
1
0
010000
0
0001000
0
1
000100
A,=
0 0 0
000010
0
0
0
0
0
0
.95
.22
0
0
0
0
self-esteem
achievement needs
locus of control
E =
goal difficulty
goal clarity
ability
.20
.27
.29
=
.19
.08
.10
.95
Because five of the observed variables (self-esteem, need for achievement, locus of control, goal difficulty,and goal clarity)were measured using multiple-itemscales, priorto the analysis the errorterms
forthese variableswere fixed at the quantityone minus their internalconsistency (reliability)multiplied
by the varianceof each measure (Hayduk,1987).This approachenables the explicitrecognitionof known
measurementerrorto be incorporatedintothe structuralequationmodel. Forexample, the internalconsistency for self-esteem was .80. The variables were standardizedto a mean of zero and a variance
of one, therefore the errorterm for self-esteem was computed as 1*(1- .80)= .20.
The remainingtwo observable variables (intellectualabilityand qualityof education) were combined
in a factormodel to estimate the conceptual variable,ability.The abilityto combine factoranalytictechniques and estimationtechniques is enabled throughthe LISRELapproachto data analysis.The squared
multiplecorrelationsof .95 and .22 (see Ax)indicatethat these two observable variables "load" on the
conceptual variableability.The factor loadings for both variables are statisticallysignificantat p<.05.
The assessment of fitof the ActualPerformanceModelto the data consists of three steps: (1) examination of the LISRELsolution, (2) measures of overallfit, and (3) detailed assessment of fit (Joreskog and
Sorbom, 1988). With respect to examinationof the solution, careful attentionwas paid to parameter
estimates, standarderrors,correlationsof parameterestimates, squared multiplecorrelations,and coefficients of determination.The main parameterestimates of interest in this study (standardizedsolution
coefficients) are all significant. In addition,there are no identificationproblems that might be caused
by highly correlated parameterestimates.
Measures of overall fit and detailed assessment of fit gave no indicationof any problems. The x2
measurewas 10.75 (p = .38),whichgave no statisticalevidence to rejectthe model.Inaddition,an analysis
of the magnitudeof standardizedresiduals, and a q-plotof standardizedresiduals did not reveal any
non-normalityor identificationproblems with the model.
This content downloaded from 134.129.120.3 on Tue, 15 Dec 2015 16:30:43 UTC
All use subject to JSTOR Terms and Conditions
413