tools for testing contingency models prof. brian boyd w. p. carey school arizona state university ...
TRANSCRIPT
Tools for Testing Contingency Models
Prof. Brian BoydW. P. Carey School
Arizona State Universityhttp://www.briankboyd.com
Kati Haynes, Mike Hitt, Don Bergh, & Dave Ketchen
Co
nti
ng
ency
Tay
lori
smRationalizationRationalization
CompartmentalizationCompartmentalization
Strategic ChoiceStrategic Choice
Confluence of multiple factorsConfluence of multiple factors
Problems With Contingency TheoryProblems With Contingency Theory
• Issues regarding causality• Imprecision regarding definition of “fit”• Near-infinite combination of variables• Limited or inconsistent support for many
theoretical predictions• Schoonhoven (1981): Contingency is an
orienting strategy or metatheory, but not a theory per se
• See Peteraf and Reed (2007) for “The rise and fall and rise again of contingency theory”
Why Are Contingency Effects Why Are Contingency Effects Important?Important?
Examples from corporate governance research:
• Meta-analyses of both duality and insider ratios report minimal links to performance
• These same variables show highly significant roles in the context of moderators, mediators, and other effects
Weak main effects in other SM areas as well:
• Strategic planning and performance
• Acquisitions and performance
• Tests of assumptions of transaction cost economics
RoadmapRoadmap
• Venkatraman’s (1989) typology as a toolbox for theory development
• Content analysis of 30 years of studies in SMJ
• Implications for future studies
Venkatraman’s Venkatraman’s (1989) Typology (1989) Typology
of Fitof Fit
LOW
FEW
MANY
HIGH
Criterion-specific Criterion-free
Nu
mb
er o
f va
riab
les
in
eq
ua
tio
n
Deg
ree
of
sp
ecif
icit
y in
fu
nct
ion
al
form
of
fit-
bas
ed
rela
tio
nsh
ip
ProfileDeviation
Moderation
Mediation
Gestalts
Matching
Covariation
Content Analysis:Content Analysis:DescriptivesDescriptives
Benchmarking Benchmarking Prior Studies Prior Studies
• Fit is central to theory development in strategic management
• First empirical article published in SMJ used a contingency design
• We examined all empirical studies published in SMJ between 1980 and 2009
• Practices/trends generally track those of other management sub-specialties
0
10
20
30
40
50
60
70
80
90A
rtic
les
/ y
ea
r
Total
Empirical
Contingency
Breakout of ArticlesBreakout of Articles
0%
20%
40%
60%
80%
100%
1980s 1990s 2000s Overall
Other
Survey
Archival
Data Sources Over TimeData Sources Over Time
0500
100015002000250030003500400045005000
Sam
ple
siz
e
Archival
Survey
Sample sizes varied widely across studies.
Gains in N are only marginally significant for both survey- and archival-based studies
Sample Size Over TimeSample Size Over Time
Conceptual
Other Empirical
Interaction
Subgroup
Mediation
Matching
Gestalt
Profile
Covariation
Articles By TypeArticles By Type
Distribution of Contingency PapersDistribution of Contingency Papers
Interaction
Subgroup
Mediation
Matching
Gestalt
Profile
Covariation
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
Interaction
Subgroup
Mediation
Matching
Gestalt
Profile
Covariation
Contingency Papers Over TimeContingency Papers Over Time
Top Theoretical PerspectivesTop Theoretical Perspectives
Rank 1980s 1990s 2000s
1 IO econ IO econ RBV
2 Contingency RBV Agency
3 SCP model Agency Knowledge
4 Org theory Contingency Networks
5 Upper echelons
TCE TCE
Content Analysis:Content Analysis:Individual ToolsIndividual Tools
ModerationModeration
Interaction:• First-order relationship
between x and y changes at different levels of z
• Arnold (1982): Form moderation
• Some potential issues:– Power– Colinearity
Subgroups:• First-order relationship
between x and y can be stronger or weaker at different levels of z
• Arnold (1982): Strength moderation
• Some potential issues:– Power– Coarse measurement of z– Less robust hypothesis
testsApproaches are interrelated, but do not always yield comparable results
Approaches are interrelated, but do not always yield comparable results
Interactions – Results Interactions – Results • Became the dominant tool by 2000s• Mainly used in regression, increasingly common in
limited dependent variables, rarely in SEM• Minority of studies distinguish between types of
moderation, stable over time• Only 20% of studies discuss power• One in three papers address measurement quality• Mean centering more widely discussed, but infrequently
used• Increasingly common to address multicolinearity issues• Interactions often not shown visually, but this is
becoming more common over time
Aguinis & Gottfredson (2010) Journal of Organizational Behavior for interaction best practices
Aguinis & Gottfredson (2010) Journal of Organizational Behavior for interaction best practices
Subgroups – Results Subgroups – Results • Second most widely used contingency tool• Comparison of correlations dramatic decline over time• Regression is primary tool, big gains in use by both SEM
and logit/probit• Minority of studies distinguish between types of
moderation, increasing number of studies reporting supplementary analyses
• Virtually none of studies discuss power• One in ten papers address measurement quality• Only 20% of studies report any significance tests of
differences between subgroups• 15 – 20% of studies use subgroups for post hoc
analyses
MediationMediation• The relationship between x and y is
affected by the intervening variable z
• Simple versus complex or multiple mediation
• Methods choices: MacKinnon et al (2002) identified 14 choices for testing effects
• Growing trend of integrating mediating and moderation effects concurrently
Wood et al (2008) ORM for a content analysis of mediation studies in micro and mixed-focus journals.
Mathieu, DeShon & Bergh (2008) ORM for an overview of mediation research designs.
Wood et al (2008) ORM for a content analysis of mediation studies in micro and mixed-focus journals.
Mathieu, DeShon & Bergh (2008) ORM for an overview of mediation research designs.
Mediation – ResultsMediation – Results• Less than five percent of studies use mediation
– comparable to other outlets• SEM is by far the dominant tool of choice, a
marked difference from the emphasis on regression in other management outlets
• Most (90%) studies use complex mediation – more than other journals
• Cross-sectional designs dominant, but (a) less so than other journals, and (b) declining over time
• Power rarely addressed, measurement more so• One third of studies consider alternate causal
configurationsMiller et al (2007) RMISM examine use of Baron and Kenny ‘four step’ in strategy studies.
Miller et al (2007) RMISM examine use of Baron and Kenny ‘four step’ in strategy studies.
Gestalts and ConfigurationsGestalts and Configurations
• Venkatraman (1989): Degree of internal coherence among a set of theoretical attributes
• Miller and Friesen (1977): Holistic and ordered patterns of attributes
• Similar to configurations and archetypes
Gestalts and Configurations – Gestalts and Configurations – Results Results
• Over time, configurations increasingly based on theory versus exploratory data analysis
• Use of multiple clustering algorithms has grown dramatically over time
• Sample size is a historical concern, and has not improved over time
• Mixed results for validity tests– Split half reliabilities and hold-out samples rarely used– Use of criterion validity tests improving
Short et al (2008) JOM review configuration studies
Short et al (2008) JOM review configuration studies
Profile DeviationProfile Deviation
• Micro-Taylorism: There is an ideal combination or bundle of elements, but that combination is highly situation dependent
• Thomas, Litschert, and Ramaswamy (1991): Firms pursuing Prospector and Defender strategies each have unique needs vis-à-vis TMT characteristics. Alignment of strategy and human capital will shape future performance
Profile Deviation – ResultsProfile Deviation – Results• Extremely rare in practice• Little consistency in defining ideal: Ranges from
10% of top performers to average firm behavior• Arguments for and against weighting of
components, but no empirical testing of the different approaches
• Multiple samples recommended, but rarely utilized
• Little consistency in approach for testing predictive validity
Zajac et al (2000) SMJ examine directional effects of departure from ideal
Zajac et al (2000) SMJ examine directional effects of departure from ideal
CovariationCovariation
• Related to gestalt, as fit is seen as a consistent pattern across a number of areas
• Venkatraman (1989): Covariation requires a high degree of theoretical precision in the linkage among elements
• Modeled as second order factor structures
• Hult and Ketchen (2004): Use the RBV to assess how four capabilities combine to create a distinct advantage for the firm
Covariation – ResultsCovariation – Results
• Also quite rarely used, but more common in recent years than profile deviation or gestalt
• Typical paper uses 4-5 first order factors, with 2-7 indicators per factor
• Few studies conduct formal comparison of first versus second order structure
• Several methodological issues– Small ratio of observations: indicators– Often no control variables included– Only partial correlation matrices reported
• Several studies have applied covariation in conjunction with mediation or moderation
MatchingMatching
• Venkatraman (1989): “A measure of fit between variables…developed independently of any performance anchor”
• Fit can be based on deviation scores, residual analysis, or three way interaction.
• Most studies in SMJ referencing “matching” were simple interaction terms
• Powell (1992) an example of profile deviation, Habib & Victor (1991) an example of three-way interaction
• Very rarely used tool
Suggestions for Suggestions for Future StudiesFuture Studies
Theory DevelopmentTheory Development
• Frame more nuanced hypotheses– Multi-theoretic perspectives– Complementary or competing, synergy or
suppression?– Combinations of different contingencies (e.g.,
moderated mediation)
• Faculty training– Declining variety observed in types of tools used in
our content analysis– Shook et al (2003: 1236): “Doctoral training may not
be keeping pace with data analytic trends and future research needs.”
MethodologyMethodology
• Design studies with sufficient power– Power is generally weak in SM studies– Broad constructs often measured with single items– Power is a concern with many contingency tools
• Continue advances with mediation– Less than 5% of studies use mediation
• Pay greater attention to type of moderator effects
• Application of interactions to limited dependent variable studies
• Rarely used tools offer great opportunities for theoretical development
Organizational Research MethodsOrganizational Research MethodsSpecial issue on
Construct Measurement in Strategic Management
Sample topics:Sample topics:• Development and validation of new
measures• Multi-level variables• Problems with proxies• Articles/notes and empirical/conceptual
papers welcomeCall for Papers in April issue of ORMCall for Papers in April issue of ORMManuscripts due December 1, 2011Manuscripts due December 1, 2011