Abstract
CEO duality describes the governance structure when a firm’s chief executive officer also holds the position of chairman of the board. Duality is central to theoretical perspectives on corporate governance and top management, yet duality’s relationship with numerous outcomes is characterized by nonsignificant coefficients and bivariate correlations hovering near zero. We argue and present evidence that CEO duality represents a “dummy construct”—an intentionally pejorative assessment on the widespread use of binomial categorical “dummy” variables to represent complex constructs. While we highlight CEO duality, the use of dummy variables as constructs is common in research. We review CEO duality as a construct and assess typical approaches to its measurement. When compared to actual patterns of duality within organizations, we find that current operationalizations are lacking due to a lack of attention to temporal considerations. This raises questions about the construct validity of current conceptualizations of CEO duality. Actual patterns suggest constructs and theoretical perspectives not previously considered. We present a taxonomy of CEO duality archetypes and offer suggestions on the incorporation of time for studies using dummy variables.
Keywords
Chief executive officer (CEO) duality—meaning when the CEO is also the chairman of the board—represents a puzzle for researchers. There are strong arguments that duality should influence a range of firm outcomes, notably firm performance (Finkelstein & D’Aveni, 1994). Arguments have been made for positive effects on governance from managerial perspectives and board leadership structure (Bourgeois & Eisenhardt, 1988; Donaldson & Davis, 1991; Eisenhardt, 1989b) and negative effects under agency perspectives (Fama & Jensen, 1983; Jensen & Meckling, 1976). Yet overall, the evidence of direct effects from either perspective is elusive. Collectively, studies report low to no relationship between CEO duality and specific organizational actions, such as earnings management (García-Meca & Sánchez-Ballesta, 2009), and firm performance overall (e.g., Boyd, Haynes, & Zona, 2011). This lack of findings has led one group of authors to observe “despite voluminous empirical attention, there is virtually no evidence related to the financial performance of the firm” (Dalton & Dalton, 2011, p. 404). They conclude: “Indeed, we are not aware of a body of literature in corporate governance—or elsewhere—where null results present with such consistency” (Dalton & Dalton, 2011, p. 408).
One potential conclusion is that no relationship exists. However, given strong theoretical basis behind expected relationships, retreating to the null seems a significant about face for the theories involved: A true lack of effect would be surprising. Several explanations have been offered to explain this lack of significant effects, including self-selection and endogeneity (Iyengar & Zampelli, 2009) and ignoring multilevel effects (Dalton & Dalton, 2011). One meta-analysis reports that all effects in studies of the relationship between CEO duality and financial reporting irregularities constitute Type I error, attributable entirely to sampling bias (García-Meca & Sánchez-Ballesta, 2009).
We propose that an alternative cause may be at play. Constructs, nondirectly observable phenomena, are common in all social sciences, including strategic management (Godfrey & Hill, 1995; Ketchen, Boyd, & Bergh 2008). Several recent reviews (e.g., Boyd, Gove, & Hitt, 2005b; Hitt, Boyd, & Li, 2004; Hitt, Gimeno, & Hoskisson, 1998) highlight measurement error for constructs in strategic management as a serious issue that can influence results (Boyd, Gove, & Hitt,, 2005a). Many constructs are poorly measured as they are represented by discrete indicators (Boyd et al., 2005b).
While measurement error is increasingly recognized, its causes are less understood. Further, scholars have begun to more clearly articulate the distinctions between measurement error and construct validity (Edwards, 2011; MacKenzie, 2003; Schwab, 1980). Of great relevance here, management theory has noted the role and importance of “time” in research generally (Bluedorn & Denhardt, 1988; George & Jones, 2000) and in strategic management specifically (Mosakowski & Earley, 2000). Time’s impact on constructs and measurement, however, is still being examined. The inclusion of time changes constructs, the conceptualized relationships between them, and their measurement (George & Jones, 2000; Mitchell & James, 2001). Theoretical constructs incorporate temporal dimensions either explicitly or through an inherent assumption of temporal stability. In many instances, this assumption is unfounded (Arundale, 1980; Bluedorn & Denhardt, 1988; Schutz, 1967).
Failure to incorporate time into theory can result in poor construct conceptualization (George & Jones, 2000; MacKenzie, 2003), resulting in the interrelated and recursive problems of poor construct validity and flawed measurement. Where construct validity is the correspondence between a conceptualized construct and the operational language used to represent it (Edwards, 2011; Schwab, 1980), a recursive relationship exists between it and measurement error. Clear constructs with solid construct validity are a prerequisite for effective measurement, which in turn is essential for the demonstration of construct validity. Untested assumptions about the temporal nature of CEO duality may result in lack of recognition of changes in duality and ignore potential cyclicality and why and how changes in duality occur within an organization. Time may affect CEO duality in many ways, including variability over time, for both predictable and unpredictable reasons. Failure to incorporate appropriate measures of a temporally unstable construct will result in measurement error and may result in assumptions of construct validity where it should be questioned. A lack of attention to the temporality of CEO duality may be one contributing cause of the nonsignificant results identified in prior work (Dalton & Dalton, 2011).
At the nadir of construct conceptualization and measurement are single binomial indicators or “dummy variables.” As measures of constructs, dummy variables represent only if something is present or not. Facets such as subdimensionality and levels are intentionally ignored in favor of parsimony. When dummy variables are used in lieu of a continuous measure, values of 0 typically represent none or the absence of something, and a 1 equals presence of something or all non-zero levels. This measurement simplification is not without costs, as the severity of errors in measuring constructs is inversely proportional to the number and sophistication of indicators used. While dummy variables may be effectively used to represent temporally stable and truly dichotomous factors and as a proxy for early stage constructs (Searle & Udell, 1970), their widespread use in strategic management research is problematic. Dummy variables represent the extreme in their reliance on temporal stability as variability within and across time cannot be incorporated. They also are at the extreme high in terms of measurement error introduced to statistical models due to this instability.
We focus our attention on CEO duality, highlighting issues with the broader use of dummy variables as representations of constructs in strategic management. Dummy variables are commonly used to represent constructs that may, knowingly or unknowingly, have greater complexity than can be proxied with a simple binomial representation. This representation raises the issue of construct validity (Schwab, 1980). Can a latent phenomenon be adequately represented by the measure? Here we highlight a commonly used dummy variable, CEO duality, to illustrate the limitations, misconceptions, and measurement issues associated with something so seemingly straightforward.
Is it a coincidence that CEO duality is among the variables in governance research commonly conceptualized and measured with a dummy variable while also persistently found to be nonsignificant? Perhaps. However, measurement error is especially problematic with categorical variables as correlations and effect sizes are more substantially reduced than in continuous variables (Arundale, 1980; Ree & Carretta, 2006). Continuous variables are rarely completely wrong, containing nothing but measurement error, but this result is often the situation of error in dummy variables. With no degrees of latitude, binomial measures force to the extreme; they are either fully correct or fully incorrect.
To better understand causes of the missing link between duality and firm outcomes, we review how duality has been measured in prior research. We focus on three aspects: time, measurement error, and construct validity. Our findings suggest that the common measure of CEO duality, cross-sectional assessment, contains a large degree of measurement error and omits important underlying processes and alternative theoretical explanations, creating model specification errors. When viewed longitudinally, possible causes of the limited relationships reported as well as alternative theoretical perspectives on duality are revealed.
We find significant variability in the use of CEO duality as a governance structure both across firms and, more importantly, within firms. The patterns of the variability appear to represent several distinct uses of duality within firms, indicating lack of construct clarity (Suddaby, 2010). We present a taxonomy of CEO duality based on these patterns of variability. Ideal types represent: stable governance structures (i.e., stable duality or nonduality), intentional changes in governance structure (i.e., switch from duality to nonduality and vice versa), and temporary changes resulting from planned or unplanned leadership transition events. We find that the current measurement practice is unable to distinguish among these approaches to duality.
Further, we find that the common practice of assessing CEO duality once based on positions held at the time of proxy statements may omit actual changes in CEO duality due to “hidden facets” (Peng & Finn, 2010, p. 368)—phenomena missed due to inappropriate temporal aggregation. This effect is problematic as, even when examined longitudinally, differences in duality are often hidden or revealed simply due to differences in proxy reporting dates.
Collectively, the findings suggest current usage of dummy variables for assessing constructs greatly reduces the field’s ability to detect intended constructs, resulting in erroneous conclusions. These limitations may help to explain why prior research involving CEO duality reports relationships and effects lower than expected by theory. We offer suggestions for improving the use of dummy variables in future research, ways they may be altered to substantially improve prediction, and alternatives to their use.
CEO Duality
CEO duality is at the heart of a diverse set of research streams, including corporate governance, top management teams, strategic leadership, and executive compensation (Finkelstein, Hambrick, & Cannella, 2009; Johnson, Daily, & Ellstrand, 1996; Zahra & Pearce, 1989). CEO duality, or its absence, is variously considered essential for agency theory (Eisenhardt, 1989a; Fama & Jensen, 1983) and other organizational theories focused on managerialism (Chahine & Tohmé, 2009; Eisenhardt, 1989b; Finkelstein & D’Aveni, 1994).
CEO duality exists, by definition, when the chief executive officer of a company simultaneously holds the title of chairman of the board of directors of the company (Finkelstein & D’Aveni, 1994; Rechner & Dalton, 1991). Nonduality finds different individuals in these two leadership capacities. This CEO duality/nonduality distinction is, however, an operational one, allowing for testing hypothesized theoretical relationships (Blalock, 1968). The constructs that duality may actually represent differ in theoretical approaches, where we accept the definition of a construct here as a conceptual term used to describe a phenomenon of theoretical interest (Edwards & Bagozzi, 2000; Schwab, 1980). CEO duality is most generally a latent construct assessing the relationship of the CEO and the board chairman.
Under the agency perspective, duality represents the power of the CEO to influence the organization in a manner inconsistent with the interest of the broader shareholder group. Duality represents combined power at the top of the organization and an enhanced ability to circumvent monitoring and engage in behaviors of self-interest. CEO duality under this view can contribute to CEO entrenchment and exacerbate already substantial agency problems, particularly in firms with the separation of ownership and control (Fama & Jensen, 1983).
Under managerial perspectives, CEO duality also represents a construct. Unlike agency’s power interpretation, here combining the positions of CEO and board chair creates a beneficial interface between the firm and environment and an internal leadership structure more readily able to respond to that environment. CEO duality represents varied information processing, decision making, and unified command and control dimensions that enhance organizational performance (Donaldson & Davis, 1991; Finkelstein & D’Aveni, 1994).
In the following sections, we detail the concept of CEO duality under two prominent managerial theories, agency and managerialism, demonstrating the centrality of CEO duality to research in these areas and the alternative conceptualizations of CEO duality.
Duality: Agency Perspective
Agency theory is well established in the management literature concerned with resolving two types of problems: when desires and goals of principal and agent conflict and when it is costly or challenging for the principal to verify agents’ actions (Eisenhardt, 1989a). The problems articulated in agency theory have taken a central position in the theory of the firm (Alchian & Demsetz, 1972; Fama, 1980; Jensen & Meckling, 1976). In a modern corporation where there is a significant separation of (shareholder) ownership and (management) control, agency theory has for some time taken an important role (Fama & Jensen, 1983).
When owners shift operational authority to managers, a costly monitoring role is necessarily introduced, without which shirking can be expected to occur. Specialized monitors may serve to reduce the shirking, but then the problem arises: Who is monitoring the monitor (Alchian & Demsetz, 1972)? The relationship between shareholders and the management of a corporation fits well the pure definition of a principal–agent relationship, and hence the separation of ownership and control in the diffusely held corporation gives rise to agency problems (Jensen & Meckling, 1976). The agent will not always act in the best interests of the principal, though the principal can invest in incentives to limit the divergences, and Jensen and Meckling (1976) define agency costs as follows: (a) monitoring expenditures by the principal, (b) bonding expenditures by the agent, and (c) residual loss.
The notions of cost and, perhaps more germane, loss are quite important. Agency problems arise because contracts are costly to write and enforce, and such costs include, as mentioned earlier, the costs of structuring, monitoring, and bonding contracts, as well as the value of the lost output (Fama & Jensen, 1983). Further, the way in which organizations manage the decision process across agents has a direct bearing on, and can help explain, the survival of the firm itself. If the CEO is also the chairman of the board (i.e., CEO duality), then monitoring is diminished. Fama and Jensen (1983) argue that under duality, decision processes are dominated by the chief executive officer, signaling “the absence of separation of decision management and decision control, and, in our theory, the organization suffers in the competition for survival” (Fama & Jensen, 1983, p. 314).
From this articulation it is straightforward to see how and why CEO duality would be seen negatively under agency theory. The principal–agent relationship between owners and managers of large and, even more so, public corporations is clear, and the agency problems along with the associated costs are very real. Such problems and costs can arguably conspire to reduce profits and performance of firms over time or even very quickly. How a firm manages these issues from the top is reflected in the decision of CEO duality or nonduality. In a modern corporation, the goal is to earn returns for its shareholders. The board of directors has a fiduciary duty to shareholders and the responsibility to appoint and fire the firm’s CEO. The CEO in turn appoints and replaces key managers when necessary. This course of action may be less likely when the CEO is also the main monitor of the organization’s management.
Duality: Managerial Perspective
Though agency theory has been a frequent perspective taken in studying governance aspects of the modern corporation, alternative perspectives exist, such as those in the organizational literature (Donaldson & Davis, 1991; Eisenhardt, 1989a). In the case of examining CEO duality, alternative perspectives to agency theory argue for different underlying mechanisms, resulting in opposing outcomes, but relying on the same operational variable.
Several authors take the position that CEO duality can provide a single focal point for leadership at the company, which can create the impression of firm stability, thereby supporting confidence in the firm’s management as well as better communication between management and the board (Iyengar & Zampelli, 2009). For example, consolidating the CEO and chairman position can serve to clarify a strong and unambiguous leadership structure (Donaldson & Davis, 1991; Finkelstein and D’Aveni, 1994). Using this argument, Finkelstein and D’Aveni (1994) examine decision-making authority and find that board vigilance is significantly positively related to CEO duality—a finding counter to agency theory predictions. The finding suggests that when other influences are held constant, vigilant boards appear to be more concerned with unity of command than with entrenchment avoidance (Finkelstein & D’Aveni, 1994). In an international context, authors have examined whether in fact the positive aspects of CEO duality may outweigh the negative agency aspects for underpricing of Arab IPOs, arguing that the focused leadership and responsiveness from CEO duality are perhaps very important to young startups in the process of going public (Chahine & Tohmé, 2009).
Eisenhardt herself has argued that in environments described as high velocity, the speed of decision making is critical and greater speeds can lead to greater performance (Eisenhardt, 1989b). This research program on strategic decision making in high-velocity environments explored the notion that in industries where the rate of technological and competitive change is so extreme, market information becomes obsolete quickly or is unavailable and speed matters (Bourgeois & Eisenhardt, 1988). Their hypotheses argue that the shorter the time frame for decision making, the better the firm performance and that effective firms vest power in top management teams for implementing strategy in high-velocity environments (Bourgeois & Eisenhardt, 1988). It appears that in certain settings, CEO duality may be a net positive for shareholder outcomes if it leads to greater speed in decision making.
Two Theories, One Dummy Variable
Together, agency and managerialism are two significant theoretical approaches in management research. Within both, CEO duality is central, yet it represents entirely different constructs. In agency studies, duality represents the loss of the separation of management and monitoring; in managerial studies, CEO duality aids decision-making speed and effectiveness. These competing benefits have long been recognized in the literature, leading CEO duality to be identified as a “double-edged sword” (Finkelstein & D’Aveni, 1994).
While the advantages and disadvantages of duality pose challenges for practical reconciliation, we see a more immediate issue with measurement. It is important to note that statistical support for these competing perspectives relies solely on differences in the directionality of a common variable. Under both perspectives, CEO duality is the independent variable, the effect of which is being assessed on a subsequent outcome. The support of theorized relationships for both rests on CEO duality’s statistical significance, but in opposing directions. Typically, a significant negative relationship with firm-level performance supports agency relationships whereas a significant positive relationship supports managerialism. We will argue that this type of relationship, when assessed with a dummy variable, is especially prone to the effects of measurement error. In the next section, we review how CEO duality is operationalized in empirical studies, focusing on the conceptualization and incorporation of time into measurement approaches, and how these choices create concerns regarding the construct.
Extant Measurement of Duality
To uncover insights as to the general lack of findings, we conducted a content analysis of empirical examinations of CEO duality in peer-reviewed outlets. We sought to identify a sufficiently large number of articles to capture the measurement approaches used. Given the large number of studies involving duality, we sought to balance the volume of studies against the need for patterns of measurement to emerge. We began with a search for articles published in the strategic management literature over the past 30 years by searching EBSCO Business Source Complete. An initial search for CEO duality and variants yielded over 450 academic articles, evidence of its importance, but an overwhelming number to examine. Therefore we narrowed the search to those in the strategic management literature using the specific phrase CEO duality. 1 We added to this review articles and meta-analyses that include duality. 2
We followed recommended practices (Duriau, Reger, & Pfarrer, 2007), identifying the journal pool based on a defined quality criteria. We used the leading strategic management outlets that publish empirical work based on MacMillan’s (1991) list. This included six journals: Academy of Management Journal, Strategic Management Journal, Journal of Management, Administrative Science Quarterly, Management Science, and Long Range Planning. We added outlets focused on strategy and corporate governance with high impact factors in the Journal Citation Report. These outlets are: Corporate Governance: An International Review, Organization Science, and Strategic Organization. These searches yielded 7 review articles and meta-analyses, 3 of which included statistical data on CEO duality, and 19 empirical studies (Table 1). While national governance contexts and cultures may result in differences in the use and perspective on CEO duality as well as its relationship with outcomes, we found the studies in the sample conducted in international contexts to take similar theoretical perspectives and report similar relationships to studies in the U.S. governance context.
CEO Duality in Empirical Studies.
Note: DV = dependent variable; IV = independent variable; NR = not reported; ROA = return on assets; ROE = return on equity; ROI = return on investment.
Within the articles identified, one utilized CEO duality as a control variable (Delgado-García, De Quevedo-Puente, & De La Fuente-Sabaté, 2010), one as a dependent variable (Finkelstein & D’Aveni, 1994), and the remainder as an independent variable. All meta-analyses examined duality as an independent variable. Consistent with expectations, the majority of studies focused on agency and/or managerial perspectives, commonly focusing on main effects. A small number of studies incorporate moderation (O’Connor, Priem, Coombs, & Gilley, 2006; Tuggle, Sirmon, Reutzel, & Bierman, 2010), contingency (Elsayed, 2007), alternative theoretical perspectives (Elsayed, 2007), or a combination of such approaches (Boyd, 1995).
We identified four pertinent reviews of corporate governance in the literature (Finegold, Benson, & Hecht, 2007; Johnson et al., 1996; Pugliese et al., 2009; Zahra & Pearce, 1989). All provide details on CEO duality or the broader topics of CEO and board relationships and top management configuration, but none detailed CEO duality measures or reported correlations between duality and other variables.
Three meta-analyses were identified that provide insight specifically on CEO duality. Rhoades, Rechner, and Sundaramurthy (2001) report an average correlation of 0.06 (range: –0.04 to 0.15) for CEO duality and outcome variables across 22 samples in 21 studies. Based on 31 studies and 69 samples, Dalton, Daily, Ellstrand, and Johnson (1998) report a smaller corrected correlation, –0.041 (range: –0.056 to 0.048), for duality and financial performance. In a meta-analysis of 35 studies of goverance characteristics and earnings management, García-Meca and Sánchez-Ballesta (2009) report a mean correlation of –0.001 between CEO and earnings management. The authors report the effects between CEO duality and financial reporting irregularities in prior studies are due solely to sampling bias (Type I error). Collectively, these meta-analyses provide evidence of both (a) the importance the literature places on CEO duality and (b) the very small relationship found between duality and outcomes within literature.
Time in Theory and Operationalizations
To assess possible measurement causes of these small effects, we examined a nascent literature on time and temporal issues in management theory. The social sciences, including management (Ancona & Chong, 1996), have long taken a static perspective toward phenomena (Arundale, 1980). “Management theorists tend to ignore the temporal boundaries of phenomena and assume invariance over key time constructs” (Suddaby, 2010, p. 349). Absent the explicit incorporation of time, management theory assumes temporal stability in both relationships and measures. This work highlights the need for greater attention to time in conceptualization (e.g., George & Jones, 2000) as well as empirical examinations (e.g., Mitchell & James, 2001). Tests of relationships between X and Y, where X is an independent and Y a dependent variable, rest on an assumption of temporal stability (Mitchell & James, 2001). Results will be attenuated or misattributed if violations of instability in X, Y, or both are not addressed.
In much of the management literature, temporal relationships are not theorized (Ancona & Chong, 1996; Bluedorn & Denhardt, 1988). Considerations such as time scale, or the size of temporal intervals, have the property of partitioning the temporal continuum into different size units, but also may be either socially constructed or objectively determined (Ancona & Chong, 1996; George & Jones, 2000; Zaheer, Albert, & Zaheer, 1999).
A lack of attention to time may be particularly problematic within strategic management as the field is about the dynamic flow or streams of actions (Mosakowski & Earley, 2000). While strategic management has been criticized for poor measurement (e.g., Boyd et al., 2005b), there is some evidence that the field incorporates time. Empirical examinations of strategic action commonly incorporate multiyear lags between causes and effects, yet the appropriateness of lag periods is more often assumed than established. How long CEO duality was in place prior to an observed performance outcome, for example, would have direct bearing on whether or not such duality conditions might impact outcomes. Ultimately, the proper inclusion of time can change the ontological meaning of a theoretical construct and relationships between constructs (George & Jones, 2000). Based on these concerns, we analyzed the measurement scheme used in the articles in our sample with particular attention to the incorporation of time.
Temporality of Measures in Duality Literature
Studies of time emphasize that unfounded temporal assumptions are problematic (Arundale, 1980; George & Jones, 2000). Instability in independent or dependent variables that is not understood will result in flawed measurement and statistical results (Mitchell & James, 2001). In our sample, the majority of studies utilized cross-sectional measurement schemes for CEO duality, a measure that does not assess temporality. Of the 19 empirical studies examined, 11 operationalized CEO duality as a single, cross-sectional binomial dummy variable. Two meta-analyses report similar usage. Of 21 studies examined by Rhoades et al. (2001), all were coded using the dummy variable approach. García-Meca and Sánchez-Ballesta (2009) indicate cross-sectional dummy coding was common in the 35 studies they examined. Overall, our review of operationalizations of CEO duality suggests the majority of these measures are cross-sectional, intentionally assuming temporal stability.
If a phenomenon is not temporally stable, the measure employed must be sufficient to capture it in a manner appropriate for testing. Several studies used methodologies that do not assume temporal stability yet do not assess the appropriateness of the aggregations utilized. As construct validity is dependent on the correspondence between a conceptual definition of a variable (construct) and the operational procedure to measure it (Schwab, 1980, p. 5), limitations in aggregation measurement will affect construct validity. Aggregation is the span and scope of phenomena that is assessed. The level of aggregation must be matched to the phenomena being examined (Schutz, 1967). Measures should incorporate appropriate sampling intervals, the span of measurement must match the span of the phenomena, and be of sufficient duration to capture the lowest frequency occurrences (Arundale, 1980). We found two general measurement approaches that did not assume temporal stability yet fell short in validating the aggregation periods used.
First, nine of the studies examined utilized duality measures or sample selection approaches that provide some incorporation of temporality. Four studies (Anlin, Lanfeng, Meilan, & Chinshun, 2007; Chen, Xuguang, & Weian, 2010; Tuggle et al., 2010; Yih-Wen & Lee, 2009) assessed duality in a partially longitudinal fashion. All measured duality cross-sectionally, but did so discretely over longitudinal periods. Each assessed CEO duality using a dummy variable annually for multiple adjacent years as part of a panel design. None report an assessment of span across periods and the 1-year aggregations were without justification. This approach limits the direct effect of measurement error for within-year relationships, but it remains for cross-year analysis. One study (Baliga, Moyer, & Rao, 1996) went beyond binary coding of the duality construct, categorizing governance structures as “pure duality,” “pure nonduality,” and “partial nonduality” categories. Multiple dummy variables were used in the statistical models, using a contrast coding approach with pure duality being the comparison group. Partial nonduality was defined as a firm having a separate CEO and chairman of the board for 1 or more, but not all, years during the 6 years studied.
A second set of studies did incorporate the notion of time span yet did not validate the periods used. O’Connor and colleagues (2006) operationalized duality as existing not just cross-sectionally, but longitudinally if duality existed at any time during a 4-year window. Two studies (Iyengar & Zampelli, 2009; Rechner & Dalton, 1991) tested the effect of duality on firm performance, comparing firms with stable duality (i.e., persistent within firm for multiple years) against firms with stable nonduality. In both cases, they incorporated the notion of multiyear stability in duality into their sample selection criteria. Both reported that samples sizes were substantially reduced from an initial pool due to the presence of temporal duality instability within firms. Neither study conducted an assessment of the sensitivity of effects under alternative aggregation periods.
Temporal Implications of Existing Measures
Overall, three quarters of studies use measures that rest on some assumption of temporal stability though it has not been demonstrated. One half of studies use a measurement scheme that ignores issues of temporality. Another quarter use panel designs and cross-sectional measurement wherein temporal instability, if incorrectly aggregated, will introduce measurement error. Of the 11 studies using a cross-sectional measurement design, 7 were published in 2007 or more recently—16 years after Rechner and Dalton (1991) raised the potential issue of instability in duality. It seems clear that the issue of poor measurement of CEO duality persists. Only one quarter of the studies utilized a multinomial measure or explicitly recognize temporal and longitudinal aspects of duality, either as a carryover effect or as a sampling procedure, yet appear to have little basis for the aggregation periods used.
Our comparison of CEO duality measures employed to the nascent literature on time in organizational theory suggests several aspects of time may not be adequately incorporated in duality studies. This deficiency raises the potential for measurement schemes that are misaligned with the temporal reality of duality within firms. This results in measurement error and affects construct validity, either of which would attenuate results and may explain the low empirical relationships reported in prior work. We argue that assumptions of temporal stability within duality may be flawed. This flaw in turn would affect the level of measurement aggregation appropriate for duality.
Understanding Duality in Organizations
To better understand if the predominant cross-sectional approach is sufficient to capture how CEO duality exists within firms, we began by reviewing firm proxy statements. We focused on companies listed on the 2010 Fortune 500 list, examining duality during the period 2000 to 2010. The individuals holding CEO and chairman titles, along with their start dates in these offices, were coded directly from proxy statements accessed through the SEC. The Fortune 500 is based on revenues, and many of the firms have legal structures such as mutual organizations and partnerships that do not require proxy filings. These firms were excluded due to lack of documents and the confounding of alternative governance structures and requirements. When proxy data were incomplete, we examined 10-K filings and Execucomp and Audit Analytics databases. Our final sample includes 457 firms with 4,582 firm years of complete duality data (i.e., data on both CEO and chairman).
Consistent with prior work, we coded duality as a dummy variable (0 = nonduality, 1 = duality) when the CEO and the chairman positions were held by the same individual as of the date of the proxy statement. It is common for leadership changes to occur at the time of a firm’s annual meeting, typically several weeks after the proxy date. We coded the office holders at the time of the proxy statement, with the proxy date as the data year. When multiple individuals were listed as office holders, such as chairman of the board, vice chairman, and chairman emeritus, we coded only the holder of the title “chairman.” In instances of co-chairmen or co-CEO, we coded both individuals as holders of the position. This coding allowed us to assess the degree of duality across firms.
Table 2 details duality within the firms. Our data appear consistent with other studies. We found duality the most common structure, averaging 68% across 2000 to 2010, declining from 73% in 2000 to 63% in 2010. This corresponds closely with prior studies using U.S. samples. O’Connor et al. (2006) report duality in 75% of firms and Tuggle et al. (2010) report 66% of firms. This result also closely matches duality rates of 77%, 71%, and 60% for 2000, 2005, and 2010 within S&P 500 companies (Spencer Stuart, 2005, 2010).
Duality by Year.
We then proceeded to assess if the cross-sectional approach fully captured the nuances of duality by examining within-firm changes in duality. For the majority of firms (379 firms, 85% of the sample), data were available for all years 2000 to 2010 (Table 3). This range allows for 10 comparison periods of duality changes (e.g., change from 2000 to 2001, 2001 to 2002, etc.). Data were not available for all firms for all years. Google, for example, ranked 102 on the 2010 Fortune list, but went public in 2004, with SEC documents first appearing in 2005. We included firms in our sample as long as data for a minimum of 3 consecutive years were available.
Detail of Comparison Periods in Sample.
To assess temporal stability, we compared duality within each firm across year-to-year periods (Table 4). On average, 11% of firms changed their duality structure across any year-to-year period. Changes ranged from a high of nearly 14% from 2000 to 2001, to a low of 9% in 2001 to 2002. Changes were not equal across firms (Table 5). Over one third of the firms, 171, had no changes in duality between 2000 and 2010. Roughly one quarter of the firms (119, 27%) changed from duality to nonduality or nonduality to duality. Another quarter (118 firms, 27%) changed, then reverted to their prior structure. Five percent of the firms (24 firms) changed duality three times, 2% (9 firms) cycled between duality and nonduality four times in the period. In 2 cases, firm’s duality met or exceeded random levels: One firm changed duality five times, another firm six.
Within-Firm Year-to-Year Changes in Duality.
aComparisons across years are all pairwise within firm for the years.
Instances of Within-Firm Duality Changes.
The variability in duality is striking, suggesting the common cross-sectional approach to measuring duality is problematic due to temporal instability. Unlike continuous or Likert-type variables where measurement error reduced reliability by a percentage, such as 20% for a measure with alpha = .80, measurement error in a dummy variable results in full inversion, coding a 1 instead of a 0, or vice versa. This is problematic in the context of tests of governance structures that are assumed to be stable and for comparative hypotheses where directional and significance combine to serve as the basis for interpretation. In our sample, if a 1-year cross-sectional measure is used, simply changing the year examined forward or backward 1 year dramatically increases the chance of measurement error. One quarter of the firms would be coded differently one out of three times. For another quarter of the firms, the chance of the coding being inverted is one in five. This appears consistent with meta-analytic analysis identifying sampling error as a significant cause of divergent findings across studies.
Changes may not occur in equal intervals but may instead be clustered, representing periods of change grouped during small periods within the sampling window. To assess this case, we examined the correlation in duality across years (Table 6). In all cases, the strongest relationship exists between duality in any one year and the adjacent years. All correlations are significant but are low when considered in the context of stability. The correlation between duality in subsequent years averages just .72. Extending the time span yields greater variability. The correlations suggest that a firm is unlikely to have the same governance structure in 2005 as existed in 2000 (r = .276) or in 2010 compared to 2005 (r = .384).
Correlation Matrix.
Note: Spearman correlations shown; listwise (N = 379 firms with complete data for all years). All correlations are significant at p < .001 (two-tailed).
Table 7 further illustrates this within-firm variability. Crosstabs of year-to-year classification of duality using pairwise criteria for 2000, 2005, and 2010 are presented in panels A, B, and C. Of 380 firms in 2000, 278 utilized duality, 102 nonduality. In 2001, 277 firms used duality and 103 nonduality. On the surface this result suggests stability, yet deeper analysis suggests instability. Of the 380 firms, 53 (14%) changed their duality structure in this 1-year period alone. Between 2000 and 2002, 69 of the 380 firms (18%) changed structure.
Cross-Tabs of Duality—Comparisons by Year
Note: Years within table indicate column headings (across table). The full sample is 457 firms based on inclusion in the 2010 Fortune 500 list. N for cross-tabs is pairwise by year combinations.
Comparing the actual patterns of CEO duality within organizations to the dominant measurement approach in the literature suggests areas of concern. Addressing these may help explain the surprisingly small associations reported between duality and dependent variables.
A Revealed Taxonomy of Duality Changes
By examining finer grained temporal aspects of the relationship between the position of CEO and chairman than have prior studies, we have identified a series of systematic patterns of change in duality structure. Figure 1 presents duality patterns we found in our examination. We forward these as a taxonomy for better understanding duality as a complex construct rather than as simply a dummy variable. We present nine unique archetype patterns of duality. It appears highly unlikely that one cross-sectional dummy variable is capable of distinguishing between these forms. Complicating the matter, as reported, CEO duality is unstable within firms, resulting in, in many instances, more than one of these archetypes occurring within a firm.
A taxonomy of duality.
The first two patterns are consistent with current views of duality: full duality and full nonduality that is stable over time. These patterns did occur within the firms examined but were most common under the unusual condition where no CEO or chairman changes occurred within the firm across the entire 2000 to 2010 period. A total of 116 firms had consistent duality in the period examined. Of those, 83 had no changes in the individual serving as the combined CEO/chairman. While this result provides stability, it raises the question as to if duality will remain when the firm undergoes a leadership change. In 29 firms, duality held stable over the course of one change in CEO/chairman, in 3 firms it held for three leaders, and in one instance the structure was stable despite four different individuals holding the dual CEO/chairman position during the examined decade.
Commitment to the nonduality structure was also evident in some firms. Fifty-six firms maintained constant separation in these roles across the years examined. Of these, nearly two-thirds, 35 of the 56, had no change in CEO. Sixteen firms maintained nonduality through two CEOs, 3 firm maintained nonduality despite three individuals holding the CEO title, and 2 firms maintained nonduality despite four CEOs. Many companies reported explicit policies on duality structure and did not deviate from it.
Two additional patterns, governance change: duality to nonduality and governance change: nonduality to duality, represent the clear and specific adoption of alternative governance structures. Consistent with normative recommendations, firms do switch from duality to nonduality. However, we also found firms that shift from nonduality to duality.
A fifth related pattern found involves firms that have a history of a duality structure but appear to suddenly deviate from this structure, only to return to it. We label this full duality with planned succession. We found this pattern of firms with planned succession events when an incumbent holding both the CEO and chairman roles resigned the CEO position while retaining the title of board chair. In most instances, the chairman title was retained for a predetermined period of time, commonly 3, 6, or 12 months from the date of CEO title transfer. The CEO was then elected chairman and duality returned. Any sample that happened to include this succession-induced gap in duality would mistakenly conclude nonduality.
These three patterns, shifts from duality to nonduality, nonduality to duality, and full duality with planned succession, are conceptually distinct yet sensitive to the operationalization used. We coded full duality with a planned succession when a firm had duality before and after a succession event, with one proxy reporting period having a shift to nonduality. In many cases, the proxy statements indicated a temporary transition period. However, it was common to find what appeared to be planned transition events and intervening periods of nonduality lasting multiple years between periods of duality. Using our 1-year transition scheme, we identified 83 instances with full duality with planned succession events, 136 instances when duality changed to nonduality, and 109 instances of firms changing from nonduality to duality.
Two additional succession patterns involving unplanned executive departures were also found. Nonduality with unplanned succession involves a firm that uses a nonduality structure but experiences an unexpected departure of the CEO, commonly through medical leave, termination, or other sudden event. The board chair, frequently the former CEO, temporarily assumes the position of CEO and chairman. Thus, duality is created, but it is atypical for the organization. We found 22 instances of this situation within our sample. Similarly, a firm that typically utilizes duality may experience unplanned succession of the combined CEO/chairman (full duality with unplanned succession). With haste, the board appoints a temporary CEO and chairman, but these offices are held by different individuals. A permanent CEO is subsequently hired and, 1 year later, elevated to the dual position of CEO/chairman, and duality is restored. This restoration occurred 12 times in our sample.
Two final patterns were identified that likely present the greatest challenge for researchers. Both involve what we label as “hidden” interruptions in an otherwise pattern of duality. All of the prior duality patterns are evidenced because the change occurs over, or across, the proxy reporting period. For example, a new CEO is hired in January of 2xx1, the firm files its proxy statement in March of 2xx1, holds its annual meeting in June of 2xx1, and the CEO is elevated to the position of board chair in January of 2xx2. For 2xx1, there exists nonduality. The change in the measure of duality is due entirely to the event crossing the reporting period. This is not always the case. We found “hidden” changes in duality, involving interruption from duality and nonduality that occurred within the organization but did not cross the proxy periods. Therefore, under common measurement approaches, no change in duality would be recognized within the firm. We label these nonduality with hidden duality during transition and duality with hidden nonduality during transition. In both cases, some succession event occurs that causes a deviation from a stable duality pattern, but the succession begins and ends within one proxy cycle. Altering the earlier example will illustrate this outcome. Assume a firm with duality has the same March 2xx1 proxy with a separate CEO and chairman in position. In June 2xx1, the CEO resigns, causing the chairman to temporarily assume the position, resulting in duality. In September, a replacement CEO is hired, and nonduality exists. In December, the CEO is elected as board chairman, and duality is restored. In the March 2xx2 proxy, duality is again reported. Duality, by all accounts, would be recorded for both 2xx1 and 2xx2 because the transition, with disruption in duality, occurred between, instead of across, proxy dates. We found 101 instances when a duality change of more than 2 months was hidden between proxy periods. As many organizational changes occur within the first 100 days a new executive is in office (Gabarro, 1987), this oversight may cover important underlying phenomena. Firms may actively use the proxy reporting dates, and the intervening time period, to consciously structure their transition periods to ensure desired reporting. No current measures of duality address such a condition.
Collectively these nine archetypes of duality greatly extend the conceptualizations of duality beyond cross-sectional approaches, with important implications for duality and the use of dummy variables to assess constructs. In several cases, existing measures are consistent with specific duality situations. However, there must be an a priori understanding of the nature of the duality within the firm to match the measure to the duality condition.
Discussion and Implications
Can a dummy variable represent a construct? Early advocates in management for the use of dummy variables to represent constructs (e.g., Searle & Udell, 1970) suggested their use as a temporary step until construct-level measures could be established. It appears their use in the context of CEO duality has long exceeded this exploratory phase. CEO duality has been measured almost exclusively using a binomial categorical dummy variable for more than two decades. It is safe to conclude that the measurement, and possibly the conceptualization, of CEO duality as a dummy construct has achieved the status of permanent fixture in the literature.
Usage continues despite a limited understanding of temporal assumptions. Our examination of CEO duality and the operationalized variables used suggests violation of these assumptions. The consequences of these violations are measurement error and questionable construct validity. Both may play a role in the limited relationships found for duality in prior studies. Overall, our findings suggest a need to alter our perspective on CEO duality from measuring what exists at a specific point in time to understanding an organization’s perspective on, usage of, and causes of change within a broader governance context. This conclusion is based on findings involving three interrelated aspects: temporal instability, omitted constructs, and aggregation issues. In this section we offer considerations for researchers specifically studying CEO duality as well as more generally for those using dummy variables.
Temporal Instability
Attention has been placed on the role of time in management theory (Arundale, 1980; George & Jones, 2000; Suddaby, 2010), yet we found little evidence of its incorporation in measures of CEO duality in the studies examined. Without including temporality into measurement design, there is an implicit assumption of temporal stability (Schwab, 1980). Our findings suggest this assumption is problematic. We find single point-in-time measures of duality neglect within-firm instability. Some firms do have stable duality structures, but the majority do not. Over the 10-year period examined, roughly two thirds had at least one change in duality with over 10% changing duality during any 1-year period. Governance structures have been conceptualized largely as enduring firm attributes, but there exists substantial instability within firms, even over short periods of time. Overall, an assumption of temporal stability does not appear valid. Under these circumstances, the use of a single dummy variable is limiting.
Those using dummy variables should recognize their inherent temporal assumptions. A basic model of a relationship between X and Y may be substantially altered with temporal variability in X or Y (Mitchell & James, 2001). If changes are present, the researcher should assess if dummy variable usage is appropriate. Several studies utilize approaches that address the influence of changes in duality when used as an independent variable. Work involving sample selection based on stability in duality (Iyengar & Zampelli, 2009; O’Connor et al., 2006; Rechner & Dalton, 1991) all overcome many of the problems of duality variability but are generalizable only to firms with stable structures. None, however, appear to acknowledge “hidden” changes in duality, and the effect of these transitions is entirely unknown.
Omitted Constructs and Model Specification Error
The error described previously may indicate systematic aspects of a previously unidentified phenomena. Failure to recognize this may result in model misspecification. The duality patterns we identified within organizations suggest duality may be unrelated to agency and managerial explanations. In many instances, changes in duality are not primary actions, but derivative. Studies involving theoretical tests of agency, managerialism, or other constructs using duality as a measure may find the strongest results using samples where duality or nonduality is not questioned. This can include firms both with and without executive turnover that clearly intended to maintain their governance structure (our Archetypes 1 and 2). It may include firms with short interruptions in structure, but what constitutes commitment to a structure despite interruption versus a change in structure is difficult to ascertain. While we have focused on instances where transitions result in changes in duality, insights may also be gained by further examination of transitions where duality is unchanged. 3 Responding to turnover of the CEO, chairman, or both as part of planned or unplanned succession appears an underappreciated theoretical and measurement aspect of duality. Harris and Helfat (1998) raise this potential, yet it appears to have been ignored in the literature.
Our search yielded no empirical studies of CEO duality that specifically examine or incorporate measures to assess succession events. Yet, it appears that changes in duality may be the result of how a firm treats succession as opposed to being the primary consideration. From our analysis it appears highly unlikely that any single dummy variable is capable of sufficiently representing such a diversity of phenomena. In contexts where alternative measures are not available, a researcher may gain benefits by isolating instances where one construct is most likely to occur. This approach reduces noise from measurement error from cases with alternative or multiple explanations. Several options are available to achieve this.
One approach is to focus on sample selection. Our archetypes of CEO duality patterns indicate nine distinct patterns of changes within firms. Two, continual duality and continual nonduality, represent truly stable structures. Selection of a sample with these effects alone based on a single dummy variable to compare these homogeneous groups may be beneficial.
Second, using multinomial measures in lieu of binomials may be beneficial. Such an approach was used in Baliga et al. (1996) representing no duality, some transitional duality, and full duality. This approach is widely used elsewhere in governance research where classification of directors was expanded to include affiliated in addition to the traditional insider and outsider distinction. Additional insights may also be gained by supplementing a dummy variable with additional dummy variables to represent other clearly dichotomizable constructs. This approach allows the creation of typologies that have long been recognized in the management literature as valuable for understanding phenomena (Doty & Glick, 1994; Fiss, 2011).
Aggregation
The use of a dummy variable combined with duality’s within-firm instability may result in several temporal aggregation errors. A cross-sectional dummy variable aggregates a firm’s history of use, changes and rationale for use, and governance state at one specific point in time into a single variable. Operational variables such as CEO duality serve as proxies for theoretical relationships that cannot themselves be observed (Blalock, 1968). The ability of a measure to serve this utility is partially limited by the degree to which the measure corresponds to the temporal length of the phenomena (Schutz, 1967).
It is safe to conclude from our analysis of duality that a single dummy variable improperly aggregates a diverse set of temporal aspects into a binomial value. We should note that aggregation is in no way conceptually separate from the issue of temporal instability and multiple constructs. We highlight aggregation span, cyclicality, and enduring states as three aggregation aspects missing from CEO duality measures. We provide avenues for incorporating these aspects into measurement design in dummy variables. Each comes with both costs and benefits that should be evaluated by the researcher.
Span
Aggregation consists of defining the appropriate measurement time span including start and end time and interval duration (George & Jones, 2000). Our review suggests duality as currently operationalized suffers several failures in aggregation: failing to capture an appropriate time span to fully assess duality or changes in it. The vast majority of CEO duality studies examined use what is perhaps best described as a measure of temporal convenience.
We found measures of duality are most often derived from annual SEC filings. All of the studies used an annual measure based on duality on a single date, the proxy date, to represent duality for 1 or more years. Measurement should follow phenomena rather than arbitrary reporting periods. Yet we find that duality changes may occur both across and between proxy periods. The presence of hidden changes in duality, patterns of succession that occur across multiple periods, indicate the existing approach to aggregation is insufficient. Isolating when changes in duality occur, rather than if duality is present at a particular arbitrary time, may provide greater insights.
The cost of data collection may rise with changes in aggregation. CEO duality may be unique as, unlike observational research where changes in a variable are unknown and therefore require continual monitoring, the additional monitoring to capture details of duality’s temporal flow appears to be modest. Changes in duality can be assessed post hoc. The same proxy statements that are used to assess duality also report dates of changes in the offices of CEOs and chairman. Exact dates are provided, so increases in data collection costs are minimal.
The applicability of the measure requires our second point: Researchers must consider longitudinal aspects of the phenomena to determine the appropriate aggregation span. Our results indicate significant within-firm variability in duality over time. The logistics of board transition processes and the timeline of proxy dates artificially influence the measure. Researchers may benefit both from examining changes in duality at a finer grained level and simultaneously considering longer term trends on organizational governance structures.
More broadly, the aggregation within both the independent variable and dependent variable within duality studies should be considered carefully. Studies of agency theory, for example, commonly examine duality’s effect on CEO compensation. Should the focus be on duality at the time of compensation? Or should the CEO duality of the prior chairman, who is negotiating a deal with a potential incoming CEO, be the focus? Or the historical trends within the firm prior to appointment? A comparison of empirical effects under alternative time aggregations may reveal insights into the appropriate temporal lags.
Cyclicality
Whereas aggregation refers to time span, cyclicality addresses recurrence. Differences in cyclicality between firms appear considerable yet unaddressed. Several aspects of cyclicality appear pertinent to CEO duality. Researchers may benefit from differentiating between firms with reoccurring changes in duality versus those with stable perspectives. Of the firms with stable governance structures in our sample, many had no changes in leadership during the period examined. For others, change in leadership and duality appeared quite regular. Recognition of time cycles and their associated duration and rates (Ancona & Chong, 1996; George & Jones, 2000) appears absent in the duality literature. None of the studies examined employed approaches that examine potential cyclicality. The causes and effects of reoccurring changes in duality may differ substantially from those where changes are infrequent.
Similarly, durations, intervals between changes of structures, and rates of change have not been incorporated into duality research. These may provide opportunities to advance our understanding of the effects of duality. While our nine archetypes focus on individual duality changes within firms, many had combinations of these types during the period examined. Lastly, interruptions to cyclicality, such as sudden versus expected succession events, may yield insights into both the causes and consequences of duality.
Enduring states
To fully capture the phenomena, researchers should consider both a firm’s duality culture and their intention in research questions. Current measures of duality are greatly influenced by a combination of the timing of leadership changes and intentional or unintended correspondence with the firm’s proxy reporting period. In many instances, firms express clear intentions of duality and nonduality length in proxy statements. In other instances there are attributions as to why a particular structure is adopted or altered. In both instances, the researcher may need to interpret the intention behind changes in structure.
The predominant cross-sectional dummy variables approach does not allow for the incorporation of causes of changes in duality. Neglecting the incorporation of time into CEO duality conceptualization and measurement is problematic because a snapshot should not be assumed to accurately reflect enduring qualities. George and Jones (2000) argue that present states may not correspond to prior and future states. Measures of duality may benefit from a shift toward assessing enduring governance aspects in lieu of a snapshot of a present state. For example, under current approaches a firm where one person has served as both CEO and chairman for the past 10 years would be coded the same as another firm with a history of nonduality interrupted by a temporary appointment of the chairman as CEO due to an unexpected executive departure. Clearly a single dummy variable fails to capture the historical trends within the firm and their overall approach to duality.
Disclosures of individuals and titles are based largely on reporting in proxy statements. Succession events may occur between proxy filing dates or be occurring during a proxy filing. Long-term trends in duality may be interrupted by the need to facilitate CEO or chairman succession. Temporary changes in duality—either duality or nonduality—may be captured because it occurs, by chance, over a proxy date. Similarly, mismatches may occur due to differences between a firm’s fiscal year and proxy dates and changes in duality within the firm. These cycles, through happenstance or intentional actions, may not become synchronized or “entrained” (Ancona & Chong, 1996). No existing measures of duality assuage this concern. While a temporary change in duality may be of great importance in some research contexts, in others, a firm’s long-term perspective may be more relevant.
Overall, these points suggest two conclusions. First, cross-sectional measures impart that a firm’s governance structure at any one point in time does not represent its historical basis. Second, where any present state is the result of prior decisions, temporality represents path dependence. Understanding such sequences is an essential aspect of theory development and testing (George & Jones, 2000). This understanding cannot be achieved using existing measures. Future work may benefit from assessing a firm’s duality culture, the long-term patterns and perspectives on a combined CEO and chairman, as well as defining what the period needed for adequately identifying such an aspect might be.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Notes
Acknowledgments
We are especially grateful for the comments and insights of ORM special issue editor Duane Ireland, our three anonymous reviewers, and Kevin Carlson. Each helped shape, guide, and improve the manuscript. Any errors or omissions that remain are solely ours.