A tale of two effects: Using longitudinal data to compare within‐ and between‐firm effects

• DOI: http://doi.org/10.1002/smj.2586
• Date: 01 July 2017
• Author: Matthew Semadeni,Michael C. Withers,S. Trevis Certo
• Published date: 01 July 2017

Strategic Management Journal

Strat. Mgmt. J.,38: 1536–1556 (2017)

Published online EarlyView 25 October 2016 in WileyOnline Library (wileyonlinelibrary.com) DOI: 10.1002/smj.2586

Received 21 November 2015;Final revisionreceived 16 August 2016

A TALE OF TWO EFFECTS: USING LONGITUDINAL

DATA TO COMPARE WITHIN- AND BETWEEN-FIRM

EFFECTS

S. TREVIS CERTO,1MICHAEL C. WITHERS,2and MATTHEW SEMADENI1,*

1W.P. Carey School of Business, Arizona State University, Tempe, Arizona, U.S.A.

2Mays Business School, Texas A&M University, College Station, Texas, U.S.A.

Research summary: We investigatethe theoretical and empirical implications of longitudinal data

in strategy research. Theoretically, longitudinal data allow strategy researchers to distinguish

between relationships among constructs within versus between rms. Empirically, longitudinal

data contain information about two types of relationships: within- and between-rm. Wedescribe

how the hybrid approach, a technique used in other disciplines, disentangles within- and

between-rm relationships. We reexamine a study of research and development expenditures to

illustrate the advantages of the hybrid approach. Based on our theory and reexamination, we

offer a series of recommendations for researchers using longitudinal data to test theoretical

perspectives.

Managerial summary: Strategy research examines two sources of variation over time: what is

occurring within the rm (e.g., Do rms perform better over time when investing more in R&D?)

and what is occurring between rms (e.g., Do rms investing more in R&D outperform rms

investing less in R&D?). These two sources may be similar or different in both direction and

magnitude, and when signicant differences exist in either direction or magnitude, researchers

must carefully consider the implication of these differences to their theoretical rationale and

statistical testing. Our article highlights the benets of theorizing and testing these two sources of

variance, providingscholars the ability to broaden both the theoretical and empirical contribution

of their research. This distinction is important to how research informs managerial decision

making. Copyright © 2016 John Wiley & Sons, Ltd.

INTRODUCTION

Since the early 2000s, strategy scholars have

increasingly relied on longitudinal data to test

theorized relationships. In 2004, approximately

15 percent of articles published in SMJ involved

longitudinal data. By 2014, more than half of

articles published in SMJ involved longitudinal

Keywords: research methods; theory testing; longitudinal

data; theory development; hybrid approach

*Correspondence to: Matthew Semadeni, PO Box 874006, Tempe

AZ 87287-4006. E-mail: semadeni@asu.edu

Copyright © 2016 John Wiley & Sons, Ltd.

data. During that period, researchers used longi-

tudinal data to test a variety of topics, including

sustainability strategies (e.g., Bansal, 2005), rm

reputation (e.g., Basdeo et al., 2006), merger waves

(e.g., Haleblian et al., 2012), strategic alliances

(e.g., Koka and Prescott, 2008), downsizing (e.g.,

Love and Nohria, 2005), and CEO compensation

(e.g., Bodolica and Spraggon, 2009).

Despite this remarkable increase, strategy

researchers have not fully capitalized on the fact

that longitudinal data are multilevel in nature.

Longitudinal data include two types of variance:

within-unit (e.g., within-rm, within-person,

Longitudinal Relationships in Strategy Research 1537

etc.) and between-unit (e.g., between-rm,

between-person, etc.) variance1. These two types of

variance correspond to two different relationships:

within-rm and between-rm relationships. This

allows researchers to consider not only whether

certain strategic behaviors across rms lead to

different performance outcomes (i.e., between-rm

effect), but also whether changes in strategic

behaviors within a rm lead to changes in the rm’s

performance (i.e., within-rm effect).

Multilevel research in organizational behav-

ior and psychology suggests that relationships

among constructs often differ across levels (Chan,

1998). Although constructs in this research often

involve individuals nested within teams, the same

issues pertain to longitudinal data (i.e., Does the

within-rm relationship between R&D and per-

formance mirror the corresponding between-rm

effect?). Cross-level isomorphism occurs when

relationships between constructs at lower lev-

els (e.g., individuals) mirror the corresponding

relationships at higher levels (e.g., teams) (Chen,

Bliese, and Mathieu, 2005; Tay, Woo, and Ver-

munt, 2014). Firebaugh (1978) points out, though,

that aggregated variables often measure different

constructs from the analogous variables at lower

levels. Accordingly, Bliese (2000: 368) suggests

that in organizational research “true isomorphism

is probably quite rare.”

If this lack of isomorphism also holds in strategic

management, many between-rm relationships

will differ in magnitude and/or direction from their

within-rm counterparts. We have considerable

evidence— in the form of published Hausman

tests— to support the premise that these rela-

tionships differ in many cases. When analyzing

longitudinal data, researchers typically invoke a

Hausman test (Hausman, 1978) to examine the

extent to which a set of relationships based on

within-rm variance (e.g., xed-effects models)

differ from a set of relationships based on a com-

bination of both within- and between-rm variance

(e.g., random-effects models). A signicant test

indicates that the two sets of relationships differ

statistically. We reviewed all longitudinal studies in

SMJ from January 2014 to October 2015 that used

1Consistent with Ronda-Pupo and Guerras-Martin’s (2012: 176)

description of the “rm” as “the object of study of strategic

management as a eld of research,” throughout the article, we

refer to within- and between-rm variance. Nevertheless, the

between versus within distinction could also apply to individuals,

groups, or industries.

a Hausman test. We found that 66percent of these

studies revealed statistically signicant differences

between the two sets of relationships.

It is this evidence of different relationships

across levels— coupled with the fact that strat-

egy scholars have largely ignored exploring these

potential differences— that motivates our work.

Unfortunately, decisions in strategy research about

which longitudinal analysis to perform are driven

largely by Hausman test results as opposed to

theory. When a Hausman test is statistically sig-

nicant, strategy researchers overwhelmingly rely

on xed-effects models. The disadvantage of this

approach, however, is that xed-effects models

simply discard all available between-rm vari-

ance. Consequently, using xed-effectsmodels pre-

vents researchers from gaining any insights about

between-rm relationships.

Through our work, we provide two contributions

to all strategy researchers, regardless of their

theoretical orientations. First, we highlight the the-

oretical implications associated with distinguishing

between within- versus between-rm relationships.

We use two of the most common theoretical

perspectives in strategy, the resource-based view

(RBV) (Barney, 1991) and agency theory (Jensen

and Meckling, 1976), to illustrate the importance

of this distinction. Our second contribution is

methodological in nature. We propose that strategy

scholars employ the hybrid approach, which is a

technique used by scholars in other disciplines to

disentangle within- versus between-rm relation-

ships when analyzing longitudinal data (Allison,

2005; Schunck, 2013). Simply stated, this technique

combines the advantages of xed-effects models

with those of random-effects models and allows

scholars to compare within- versus between-rm

relationships among constructs.

After explaining the intuition underlying this

technique, we reconsider Chen and Miller’s (2007)

investigation of R&D intensity to show how the

hybrid approach provides information above and

beyond a traditional xed-effects model. Our

analyses demonstrate substantive differences in

within- versus between-rm effects. As Bliese

(2000) points out, a within-rm effect that sta-

tistically differs from a between-rm effect does

not imply that either effect is “wrong.” Instead,

such ndings indicate only that scholars cannot

generalize an estimate of the within-rm effect to

the between-rm effect and vice versa.

Copyright © 2016 John Wiley & Sons, Ltd. Strat. Mgmt. J.,38: 1536–1556 (2017)

DOI: 10.1002/smj