Faults and Faultlines: The Effects of Board Faultlines on CEO Dismissal

Firm Performance and CEO Dismissal

A firm’s financial performance is a major predictor of CEO dismissal. Boards of directors’ primary responsibilities include executive personnel decisions, such as CEO dismissal. When evaluating the CEO, boards rely less on an absolute level of performance and more on performance relative to appropriate reference points (Gibbons & Murphy, 1990; Holmström, 1982). Directors establish expectations and aspirations about the appropriate level of financial performance based on similar firms as well as the performance history of the firm itself (Greve, 1998; Puffer & Weintrop, 1991). They closely monitor metrics of firm performance relative to social and historical reference points when making decisions about executive compensation, retention, and dismissal (Fredrickson et al., 1988; Jenter & Kanaan, 2015). When firm performance declines below aspiration levels, it captures the board’s attention. Directors may consider various remedies to restore profitability and appease stakeholders. If firm performance deteriorates below the aspiration level yet incremental remedies under current leadership are deemed ineffective, directors may replace the CEO to initiate drastic changes and restore stakeholder confidence. Indeed, decades of research indicate that poor firm performance is the most significant predictor of CEO dismissal (Furtado & Karan, 1990; Hilger et al., 2013; Jenter & Lewellen, 2021).

However, the empirical link between firm performance and CEO dismissal is surprisingly tenuous. In an early review, Fredrickson et al. (1988: 255-256) noted that “organizational performance explained less than one-half the variance in CEO dismissal and turnover rates.” According to a recent meta-analysis, while many studies in the literature found a negative relationship between firm performance and dismissal, the relationship is inconclusive: “The extent to which poor firm performance is related to executive dismissal depends on both the context of the dismissal and the selected performance measures.” (Hilger et al., 2013: 16)

Many factors moderate the relationship between firm performance and CEO dismissal. For example, at the level of individual executives, CEOs’ power can protect them from dismissal, weakening the relationship between poor firm performance and CEO dismissal (Boeker, 1992; Flickinger, Wrage, Tuschke, & Bresser, 2016; Shen & Cannella, 2002). At the board level, boards that consist of a greater proportion of independent outsiders tend to respond readily to low firm performance by dismissing the CEO (Gregory-Smith, Thompson, & Wright, 2009; Weisbach, 1988). In the present study, we propose that group dynamics of the board can moderate the relationship between firm performance and CEO dismissal. The board of directors is ultimately responsible for the decision to fire a CEO, and the board has to make this critical decision as a group. Typically, a sitting CEO has significant sociopolitical power over other corporate constituencies, including the board, and is usually savvy enough to take actions to circumvent any threats to the top position (Boeker, 1992; Cannella & Shen, 2001). When firm performance declines below expectations or aspirations, attempts to overthrow the CEO require unity, coordination, and decisiveness of the board (Haleblian & Rajagopalan, 2006; Weber & Wiersema, 2017). Conflicts and disintegration among board members can potentially foil a board’s attempt to take decisive action. On the other hand, exchange of diverse perspectives and expertise among directors may improve a board’s ability to process available information and assess its CEO more accurately, which can contribute to a board’s ability to make a prudent decision about how to respond to low performance. While individual director attributes and board composition are important, group dynamics affects the ability of the directors to interpret available information, find appropriate responses, and take decisive action as a group (Forbes & Milliken, 1999). In the next section, we examine an important group-level attribute of boards—faultlines—as a moderator of the relationship between firm performance and CEO dismissal.

Moderating Effect of Board Faultlines

Diversity in corporate boards has attracted much attention among corporate governance researchers (Guldiken, Mallon, Fainshmidt, & Judge, 2019; Knippen, Shen, & Zhu, 2019). The concept of diversity in teams typically refers to the degree of dispersion, dissimilarity, or heterogeneity along a single individual attribute of the team members, such as age, gender, race, or functional background (Bell, Villado, Lukasik, Belau, & Briggs, 2011; Harrison & Klein, 2007). While diversity in teams is an important concept in both academic research and business practice, the existing approach to studying diversity in teams has two important limitations. First, diversity research typically measures individual members’ attributes and then aggregates them up to the team level. Such aggregates are often the basis of measures that represent a team’s social characteristics—such as cohesion, conflicts, cooperation, and communication within the team. Group dynamics are reduced to differences among individuals. While this approach is pragmatic, it is based on an ontological assumption that a team is merely an aggregation of individuals, which stops short of the research goal of studying social dynamics within a team. The second limitation is that diversity research examines a single attribute of individuals in isolation or in an additive way (van Knippenberg & Schippers, 2007). Even when multiple attributes are simultaneously considered to construct a composite measure, for example, the demographic heterogeneity of a team, the calculation inevitably considers one attribute at a time before being added to other attributes. In this approach, the overall configuration or structure of multiple attributes cannot be directly examined.

The faultline approach aims to overcome such limitations by examining the alignment of multiple attributes simultaneously. In general, individuals have multiple attributes that constitute the perception of social identity. Individuals use these attributes to categorize themselves and others into different groups. To illustrate, consider two hypothetical teams: A and B. Team A has two White men with a finance background and two Black women with a marketing background. Team B, which also has four members, comprises one White man with a finance background, one White woman with a marketing background, one Black man with a finance background, and one Black woman with a marketing background. Although the two teams have the same degree of diversity when considering one attribute at a time (i.e., 50% women, 50% Black, and 50% marketers), when the three attributes (i.e., gender, race, and functional background) are considered simultaneously, team A has more pronounced subgroups (i.e., two White male financiers and two non-White female marketers) because all three dimensions are aligned. In team B, none of the two or more members align on more than two attributes, and thus it is unlikely that any subgroup will form based on the three attributes. In this example, therefore, team A has a stronger faultline than team B. A conventional way to examine group diversity by considering a single attribute (for example, gender, race, or age) separately would not correctly capture the degree of actual differences among members. Overall, given that individuals in group settings rarely perceive each other solely along a single attribute, but often perceive others as a joint set of multiple attributes, a faultline approach enables us to conceptualize more realistic perceptions about self and others, which form a basis of identity, social categories, and interpersonal relations (Lau & Murnighan, 1998; van Knippenberg & Schippers, 2007).

Faultline research recognizes that the effect of faultlines on group performance may not be uniform. Studies have found that faultlines tend to increase conflicts within teams, reduce satisfaction among team members, and ultimately, decrease team performance (Bezrukova, Thatcher, Jehn, & Spell, 2012; Li & Hambrick, 2005; Thatcher, Jehn, & Zanutto, 2003). At the same time, some studies have found positive effects of faultlines on innovation, learning, and performance (Gibson & Vermeulen, 2003; Hutzschenreuter & Horstkotte, 2013). The mixed findings suggest that the effect may depend on the type, or nature, of the individual attributes that constitute the faultlines (Bezrukova et al., 2009). Whereas studies in the team diversity literature typically focus on demographic attributes, such as age, gender, and race (Bell et al., 2011), boards of directors offer a unique context where nondemographic attributes also matter. Specifically, whether a director is an insider or outsider has long been considered a key aspect of board composition in corporate governance research (Johnson, Daily, & Ellstrand, 1996). The insider-versus-outsider distinction may function as a salient attribute that can potentially split the board into subgroups. Relatedly, outside directors have various job titles, such as executive manager, financier, or lawyer. It is therefore important to consider primary job titles of directors as individual attributes that reflect heterogeneous perspectives and identities, thereby creating potential faultlines (Hillman, Cannella, & Paetzold, 2000; Hillman, Nicholson, & Shropshire, 2008). Finally, a director’s tenure on the board is an important attribute that reflects their cumulative board experience, the degree of social cohesion among directors, and their sense of loyalty to the CEO (Veltrop, Molleman, Hooghiemstra, & van Ees, 2018; Westphal, 1999). Jointly addressing these disparate attributes of directors can deepen our understanding of interpersonal relations within the board.

We therefore test separate hypotheses regarding two types of faultlines: demographic faultlines (based on age, gender, and race) and information-related faultlines (based on director independence, primary job title, and tenure on the focal board). Demographic faultlines, also referred to as social category faultlines or relationship-oriented faultlines, are based on salient phenotypic attributes of individuals and may therefore contribute to relationship conflict among members (Choi & Sy, 2010). Information-related faultlines, also referred to as knowledge-based faultlines or task-related faultlines, are based on skills, knowledge, and perspectives that typically stem from educational background, training, and work experience and can potentially lead to task-related conflict (Jehn, Northcraft, & Neale, 1999). Although the two types may have some overlaps—for example, the demographic background of an individual can influence their life opportunities and work experiences—we argue that demographic faultlines and information-related faultlines are distinct constructs that must be examined separately.

In the boards-of-directors literature, few attempts have been made to examine both types of faultlines separately in a single study. Veltrop et al. (2015) examined demographic faultlines based on the age and gender of board members; Kaczmarek et al. (2012) analyzed task-related faultlines based on the type of directorship (executive vs. nonexecutive directors), education, board tenure, and financial background. In a study of board attention to entrepreneurial issues, Tuggle et al. (2010) examined overall faultlines based on board tenure, functional background, and firm/industry background of directors without contrasting the different types of faultlines. In the present study, we advance the board literature by separately testing the effects of two types of faultlines: faultlines based on demographic attributes (“demographic faultlines”) and faultlines based on information-related attributes (“information-related faultlines”).

Data and Method

We used data from the boards of directors of S&P 500 firms from 1998 to 2013. To construct the board sample, we first started with 500 firms in the S&P 500 index in each of the years from 1998 to 2013. We obtained valid data from ExecuComp for 612 unique firms, 1,147 unique CEOs, and 825 CEO departure events. We then matched the firm sample with board-related data from the Institutional Shareholder Services (ISS) Directors data set and proxy statements. After matching, we had 13,114 unique directors and 81,171 director-year observations. For the succession events, we manually coded CEO departure types using information from news reports in ProQuest ABI/Inform database. We matched the sample with company financial and operational data from ExecuComp and other sources, including Mergent Online, company annual reports, and various corporate directories. Our final sample included 552 unique firms, 1,057 unique CEOs, 12,573 unique directors, and 577 CEO departures.

Independent Variables

The primary independent variable is the firm’s financial performance relative to aspiration levels. The board’s evaluation of the CEO is rarely based on absolute performance of the firm on its own. Performance relative to aspirations is directly relevant to the board’s decision to dismiss the CEO, as suggested both by the performance feedback research and agency-theoretic view of relative performance evaluation (Greve, 1998; Jenter & Kanaan, 2015; Puffer & Weintrop, 1991).

We distinguish two types of aspiration levels: social and historical aspiration levels (Greve, 1998). Social aspiration levels are determined by the performance of firms that are deemed similar to the focal firms. We used the median return on assets (ROA) of all firms in ExecuComp that are in the same two-digit Standard Industrial Classification (SIC) code as the focal firm in the same year, excluding the focal firm. We used ROA, an accounting-based measure, rather than a market-based measure, such as shareholder returns, because our results suggest that ROA was a stronger predictor of CEO dismissal compared with shareholder returns. This is reflected in a review article, which stated that “much of this literature relies on accounting-based measures of firm performance” (Hilger et al., 2013: 16). We also controlled for total shareholder returns in all models. We used industry median ROA to minimize the impact of outliers. Results using industry mean ROA are presented in Appendix Table A2.

Historical aspiration levels are determined by the performance history of the firm itself (Greve, 1998; Levinthal & March, 1981). We calculated the historical aspiration level using the exponentially weighted moving average of experienced performance, as defined in Greve (1998) and Levinthal and March (1981):

A t = α P t − 1 + ( 1 − α ) A t − 1 ,

where A denotes aspiration level, P denotes performance measured in ROA, t denotes time period, and α denotes an adjustment parameter. Following Zhang and Greve (2019), we used 0.9 as the value of the adjustment parameter α.

The independent variable, the firm’s financial performance relative to aspiration levels, is defined as the difference between the focal firm’s performance and the aspiration levels (Greve, 1998). Specifically, ROA relative to social aspiration level is the difference between the focal firm’s ROA and the social aspiration level of ROA, as defined previously. ROA relative to historical aspiration level is the difference between the focal firm’s ROA and the historical aspiration level of ROA, as defined previously.

Moderating Variables

We test the moderating effect of the strength of board faultlines within each board. We measured demographic faultline strength by using three demographic attributes: age, gender, and race. We included age, gender (coded 1 for female and 0 for male), and race (White, African American, Asian, and Hispanic) because they are the most salient demographic attributes as a basis of social identity and self-categorization processes (Harrison, Price, & Bell, 1998; Tsui, Egan, & O'Reilly, 1992). In the ISS Directors data, 1,781 out of 12,573 directors (14.1% of the sample) had missing values on race. There is a challenge in manually coding directors’ race. Information about director race—such as photographs and other information that provides evidence (e.g., education background and awards from ethnic affinity groups)—is often unavailable for many directors. Furthermore, even when such information is found in proxy statements or public sources, relying on human judgment to determine a director’s race is prone to biases and errors. Instead, we used sociolinguistic computer algorithm developed by NamSor.com, which utilizes the directors’ full names to categorize their races. ⁴ Among the 1,781 directors with missing race data, 75 were known by their initials only instead of first names, and therefore the NamSor algorithm was unable to determine race. We used estimated race data for the remaining 1,706 directors. When we did not use the estimated race data and instead kept the race information for those directors as missing, the results (available upon request) were similar.

Information-related faultline strength was calculated by using three attributes related to the director’s task-related knowledge, information, and perspective: director independence, director’s primary job title outside the focal firm, and director tenure on the board. Differences between inside directors and outside directors may affect how they perceive themselves and others within the board. For example, the career and livelihood of insiders are more closely related to the CEO and to the firm than those of outsiders. Also, unlike outsiders who have their primary jobs elsewhere, insiders work at the same firm on a daily basis as members of the senior executive team. These fundamental distinctions may create a potential faultline between insiders and outsiders.

We also included directors’ primary job title because it reflects each director’s identity and perspectives, which may be another source of potential faultlines (Hillman et al., 2008). Directors may categorize other directors as in-group or out-group members based on their primary job title. For example, directors who are CEOs of other firms may perceive themselves as the same group (i.e., CEO group) while viewing non-CEO directors (e.g., financiers, lawyers, and professors) as out-group members. Based on directors’ primary job titles, we classified them into four groups that Hillman et al. (2000) proposed: (a) insiders, (b) business experts, (c) support specialists, and (d) community influentials. This classification scheme is based on the nature of the environmental linkages that the directors make: provision of specific resources, channels for communicating information, aids in obtaining commitments or support, and maintenance of organizational legitimacy (Pfeffer & Salancik, 1978). We combined “insiders” and “business experts” into a single category because the insider category overlaps with the director independence variable that we already use as another attribute of directors. Director tenure was measured as the total number of years since being appointed as a director.

To calculate faultline strength, we used the average silhouette width (ASW) clustering procedure (Meyer & Glenz, 2013). The algorithm computes the strength of the faultlines that can potentially split a team into multiple subgroups, which is consistent with the original concept proposed by Lau and Murnighan (1998) and Thatcher et al. (2003). Compared with several other measures of faultline strengths, including the Fau measure (Thatcher et al., 2003), the ASW has several important advantages (Meyer & Glenz, 2013; Meyer, Glenz, Antino, Rico, & González-Romá, 2014). First, unlike the most frequently used Fau measure (Thatcher et al., 2003), the ASW does not assume that there are only two subgroups within a team and is therefore suitable for teams that can potentially have more than two subgroups. In our sample, boards have 10.7 members on average, with a maximum of 27. It is unrealistic to assume that boards necessarily comprise only two subgroups. Second, the ASW is suitable for both small and large groups, unlike other measures that are suitable only if the group size does not exceed 10 (Carton & Cummings, 2013). Third, the ASW can be computed based on both categorical and numeric attributes, unlike some measures that accommodate only categorical attributes (Shaw, 2004). Our director data include both categorical (e.g., race and job titles) and numeric (e.g., age and tenure) attributes. Finally, the ASW measures the same underlying construct as the Fau (Thatcher et al., 2003) by identifying the configuration of clusters (i.e., subgroups) that have maximum internal homogeneity and maximum external heterogeneity. A simulation analysis by Meyer and Glenz (2013) shows that the ASW exhibits high predictive validity and is robust against missing values.

The calculation of the ASW has two steps. In the first step, two commonly used clustering procedures—average linkage and Ward clustering—are used to identify all possible configurations of clusters (i.e., small groups) within each group. In the second step, the ASW score, defined next, is calculated for each member and each of all possible configurations identified in the first step. The ASW is the average of each member’s individual silhouette width, which denotes how well a team member fits into their subgroup. Formally, an individual i’s silhouette width s is defined as

s ( i ) = b i − a i max ( a i , b i ) ,

where a_i denotes the average dissimilarity of i to all members of cluster A, and b_i denotes the average dissimilarity of i and all members of cluster B. The Euclidean distance is used as a measure of dissimilarity. ⁵ Categorical attributes are dummy coded by assigning a value of 1/√2 for those who belong to a given category and 0 for those who do not, following Thatcher et al. (2003). In line with Thatcher et al.’s (2003) approach, director age was given a weight of 0.1, which means that an age difference of 10 years is equally weighted as a difference in categorical attributes (for example, men vs. women). Among directors in large firms, the distribution of tenure is not very dispersed, which suggests a need for a finer distinction among the directors. We therefore used a weight of 0.2, which means that a tenure difference of 5 years is equally weighted as a difference in categorical attributes. All categorical attributes are equally weighted (i.e., a weight of 1). For a given team, the subgroup configuration that has the highest ASW value is selected, and its ASW value is used as a measure of faultline strength. The ASW value ranges between −1 and 1, where 1 represents the case where all subgroups are perfectly homogenous and faultlines are maximally strong. See Meyer and Glenz (2013) for details and illustrative example of the calculation. Studies that use the ASW include Meyer, Shemla, Li, and Wegge (2015); Mo, Ling, and Xie (2019); and Zhang and Greve (2019).

Control Variables

Firm size is known as a strong predictor of CEO dismissal (Fredrickson et al., 1988; Hilger et al., 2013). We controlled for firm sales as a proxy for firm size, which was measured as the natural logarithm of total firm sales.

The models control for several characteristics of the CEO. Older CEOs are more likely to retire than to get fired (Peters & Wagner, 2014); we controlled for CEO age in years. Female CEOs are more likely to be dismissed than their male counterparts (Gupta, Mortal, Silveri, Sun, & Turban, 2020); the variable female CEO was coded as 1 for female CEOs and 0 for male CEOs. Outsider CEOs may be at a greater risk of being fired compared with insider CEOs (Shen & Cannella, 2002); outside CEO is a binary variable coded 1 for CEOs who had firm tenure of less than 2 years when they assumed the position (Zhang & Rajagopalan, 2010).

As power dynamics between the CEO and the board may affect CEO dismissal (Boeker, 1992; Shen & Cannella, 2002), we controlled for six indicators of CEO power, director power, and CEO–board relative power. First, CEOs who hold the position of the board chair tend to be more powerful and less susceptible to dismissal (Cannella & Lubatkin, 1993); CEO–chair duality is a binary variable coded as 1 for CEO-chairs and 0 for others. Second, ownership of the firm’s shares can provide power and control (Finkelstein, 1992); CEO ownership is the percentage of the firm’s outstanding shares held by the incumbent CEO. Third, stock ownership held by the directors can bring power to the board (Cannella & Shen, 2001); we controlled for director ownership, measured as the percentage of the firm’s outstanding shares held by the directors of the firm. Fourth, a director’s tenure on the board represents the accumulation of firm-specific human and social capital and informational advantages, which may contribute to their power (Krause, Withers, & Semadeni, 2017); we controlled for average director tenure, measured in years. Fifth, we controlled for the proportion of directors appointed after the CEO took office because such directors may feel loyal to the CEO, who typically had an influence over the director appointment (Westphal & Zajac, 1995). Finally, directors who sit on many other boards outside the focal firm may have a high level of social capital and informal power (Flickinger et al., 2016); average director external directorships was measured as the average number of director positions outside the focal firm held by outside directors (Zhang, 2008).

Because CEO dismissal is largely driven by the board (Haleblian & Rajagopalan, 2006; Zhang, 2008), we controlled for board characteristics relevant to board decision-making. Board size was measured as the total number of directors on each board. Demographic similarity between the CEO and the board members may create in-group bias, which may reduce the likelihood of CEO firing (Westphal & Zajac, 1995; Zhu & Westphal, 2014). We considered CEO–board similarity in terms of age, gender, and race. For age similarity, we calculated the Euclidean distance between the CEO’s age and the average age of board members (Wagner, Pfeffer, & O’Reilly, 1984; Westphal & Zajac, 1995) and subtracted each firm’s score from the highest value in the sample to create a measure of similarity. For gender and race similarity measures, we calculated a squared value of the proportion of CEO–director dyads sharing the same demographic category (i.e., same gender or same race between the CEO and a director, based on the four racial categories used in the faultlines measures). We standardized and took the average of the three (i.e., age, gender, and race) similarity measures to create an overall similarity variable.

To assess the distinct effect of faultlines that cannot be captured by traditional diversity measures, we controlled for six measures of board diversity, one for each of the director individual attributes. Director age heterogeneity was defined as the coefficient of variation of ages among all directors on a given board. Proportion of female directors was defined as the proportion of female directors to all directors on a given board. Director racial heterogeneity was measured using an inverse value of the Herfindahl-Hirschman index, which indicates whether directors’ race is concentrated in a specific category or distributed evenly across different race categories. We used the same four racial categories used in the faultline measures. Proportion of independent directors was measured as the proportion of directors who are independent from the firm. Director job title heterogeneity was measured using an inverse value of the Herfindahl-Hirschman index. We used a modified version of Hillman et al.’s (2000) classification: (a) insiders and business experts, (b) support specialists, and (c) community influentials. Director tenure heterogeneity was measured as the coefficient of variation of the director’s tenure on the board.

All models include dummy variables for year and two-digit SIC industries. Coefficients for the year dummies and industry dummies are omitted in the regression tables. All independent, moderator, and control variables are lagged by 1 year to facilitate causal inference. All independent and moderator variables are mean-centered before entering into the regression models. Table 1 presents the descriptive statistics and correlation matrix for the variables included in the analysis.

OPEN IN VIEWER

Table 1

Descriptive Statistics and Correlation Matrix

Variable	M	SD	1	2	3	4	5	6	7	8	9	10	11	12	13	14
1. CEO dismissal	0.03	0.16
2. Log of total firm sales	8.81	1.27	.04
3. CEO age	56.29	6.41	−.01	.09
4. Female CEO	0.02	0.14	.01	.04	−.06
5. Outside CEO	0.19	0.39	.04	−.02	−.07	.00
6. CEO–chair duality	0.59	0.49	−.02	.10	.22	−.05	−.05
7. CEO ownership	1.51	5.87	−.03	−.14	.03	−.03	−.09	.04
8. Average director ownership	0.83	2.87	−.02	−.14	−.06	−.03	−.04	−.04	.54
9. Average director tenure	9.51	3.38	−.03	−.11	.13	−.07	−.19	−.10	.15	.20
10. Proportion directors after CEO	0.50	0.31	−.01	−.12	.24	−.01	−.05	.12	.17	.02	−.18
11. Average director external directorships	1.11	0.56	.00	.35	.02	.01	.04	.18	−.12	−.14	−.22	−.11
12. Board size	10.68	2.54	.03	.44	.10	.00	−.03	.08	−.11	−.12	−.08	−.12	.18
13. CEO–board similarity	0.01	0.56	−.01	−.18	.37	−.36	−.03	.10	.06	.01	.06	.13	−.07	.01
14. Director age heterogeneity	0.12	0.04	−.01	−.21	−.29	−.03	−.07	−.16	.09	.18	.14	.06	−.27	−.11	−.26
15. Proportion female directors	0.14	0.09	.05	.30	−.03	.25	.04	.03	−.10	−.11	−.14	−.11	.17	.19	−.57	−.09
16. Director racial heterogeneity	0.17	0.15	.01	.37	.01	.10	.02	.07	−.11	−.12	−.18	−.04	.24	.21	−.50	−.13
17. Proportion independent directors	0.75	0.15	.01	.19	.05	.04	.11	.08	−.23	−.32	−.28	−.01	.27	.05	−.13	−.34
18. Director job title heterogeneity	0.37	0.16	.01	−.10	.03	.04	.05	−.11	.00	.02	−.01	.07	−.18	−.05	−.03	.06
19. Director tenure heterogeneity	0.66	0.23	.03	.07	−.01	−.01	.03	−.03	.01	.05	.22	−.14	−.08	.13	−.03	.16
20. Shareholder returns (social aspiration)	0.00	0.99	−.06	−.09	−.02	−.02	.02	−.01	.08	.04	−.04	.07	−.04	−.10	.02	.04
21. Shareholder returns (historical aspiration)	0.00	1.00	−.02	−.04	−.02	.00	.00	−.03	.05	.03	−.02	.01	.00	−.04	−.01	.01
22. ROA (social aspiration)	0.00	1.00	−.04	.03	.00	.02	−.05	−.01	.03	.02	.08	−.05	−.02	−.07	−.01	−.02
23. ROA (historical aspiration)	0.00	1.00	−.09	−.02	.01	.00	−.02	−.04	.02	.03	.03	−.02	.00	−.04	−.01	.01
24. Demographic fault line	0.00	0.08	.00	.01	−.04	−.04	−.03	−.06	−.05	−.04	−.01	.00	−.03	.09	.01	.12
25. Information-related fault line	0.00	0.09	.00	.07	.02	−.03	−.02	−.05	.06	.07	.06	.07	−.03	.14	−.01	.10
26. ROA (Social Aspiration) × Demographic Fault Line	0.00	0.10	.00	.03	−.04	−.02	.00	−.02	.03	.02	.00	−.06	−.01	.03	−.01	.01
27. ROA (Historical Aspiration) × Demographic Fault Line	0.00	0.09	.04	.00	−.02	−.02	.00	.01	−.03	.00	.02	−.04	−.04	−.01	−.01	.03
28. ROA (Social Aspiration) × Information-Related Fault Line	0.00	0.12	−.03	−.04	−.05	−.01	.00	−.01	.06	.05	.02	.01	−.05	−.04	−.03	.09
29. ROA (Historical Aspiration) × Information-Related Fault Line	0.00	0.11	.02	−.03	−.03	.00	.02	−.01	.01	.03	.01	.01	−.02	−.02	−.02	.03
			15	16	17	18	19	20	21	22	23	24	25	26	27	28
16. Director racial heterogeneity			.29
17. Proportion independent directors			.25	.23
18. Director job title heterogeneity			.03	−.08	−.01
19. Director tenure heterogeneity			.03	−.01	−.09	.00
20. Shareholder returns (social aspiration)			−.05	−.04	−.03	.03	−.04
21. Shareholder returns (historical aspiration)			.00	.00	.01	.03	−.03	−.01
22. ROA (social aspiration)			.02	.02	−.05	−.03	.00	−.01	.00
23. ROA (historical aspiration)			.00	.01	−.01	−.01	−.01	−.01	−.01	.00
24. Demographic fault line			−.03	−.06	−.03	.03	.03	−.01	.03	−.03	.00
25. Information-related fault line			.03	.00	−.08	−.03	.19	−.02	−.01	.02	.01	.12
26. ROA (Social Aspiration) × Demographic Fault Line			.02	.00	−.01	.00	.02	−.02	−.01	.11	−.04	.07	.04
27. ROA (Historical Aspiration) × Demographic Fault Line			−.01	−.01	−.02	−.01	.00	.04	.04	−.04	.04	.00	.00	−.01
28. ROA (Social Aspiration) × Information-Related Fault Line			−.01	−.05	−.04	.01	.02	.05	−.04	−.13	−.05	.04	.01	.07	.03
29. ROA (Historical Aspiration) × Information-Related Fault Line			−.02	−.01	−.03	−.01	−.01	.00	.03	−.05	−.02	.00	.00	.04	.11	.01

Note: N = 5,951 firm-year observations. All correlations .026 or greater are significant at p < .05. ROA = return on assets.

Analysis

We tested the hypotheses using a continuous-time event history analysis estimated by the Cox proportional hazards regression model. As a semiparametric model, the Cox proportional hazard model has an advantage of not assuming a specific functional form of the baseline hazard (Cleves, Gould, & Gutierrez, 2010). The time clock for CEO dismissal is the tenure of the incumbent CEO. The unit of time is in months, providing improved precision over other studies that used annual data. Using year as the unit of time yields similar results (see Appendix Table A3). Following Ocasio (1994) and Shen and Cannella (2002), we split CEO tenure into fiscal years to allow time-varying covariates. All independent, moderator, and control variables are updated annually.

In the Cox model, the hazard rate of dismissal for the ith individual is

h ( t | x j ) = h 0 ( t ) exp ( x j β x ) ,

where h₀(t) is the baseline hazard function under the standard conditions, x _j  = 0, and β _x demonstrates the effects of independent variables x _j on the hazard of CEO dismissal. A postestimation test using Schoenfeld residuals indicates that the proportional hazards assumption of the Cox regression is not violated (chi-square = 87.40, df = 76, p > .05). We also performed a link test (Cleves et al., 2010) to check the proportional hazard assumption. The results indicated that the coefficient of the squared linear predictor is not significant (p > .05), which means that the proportional hazard assumption is not violated. As robustness checks, we also estimated piecewise exponential models and discrete-time event history models; the results were substantively similar to the Cox regression results. We used the Breslow method (Breslow, 1974) for handling tied events (i.e., multiple CEO dismissal events in a given month) for our main analysis. When we used the Efron method (Efron, 1977) instead, the results were similar. Among the 1,057 CEOs included in the analysis, 14 had two separate CEO job spells; for example, James McNerney was a CEO at 3M (2001-2005) and then at Boeing (2005-2015). When we excluded the 14 CEOs and reestimated the models, the findings were unchanged. These results are available upon request.

Results

Table 2 reports the Cox regression models predicting the hazard rates of CEO dismissal. Model 1 is a baseline model with control variables only. The results are generally consistent with the existing research about CEO dismissal. CEOs at larger firms, CEOs at financially struggling firms, and outside CEOs have a higher hazard of dismissal. Model 2 includes the demographic faultlines variable, which is negative but not statistically significant. Model 3 tests Hypothesis 1, which predicted that the effect of firm performance below the aspiration level is weakened when board demographic faultlines are stronger. To test the hypothesis, we interact demographic faultlines variable with two types of firm performance: ROA relative to social aspiration levels and ROA relative to historical aspiration levels. As directors may simultaneously consider both types of aspirations when evaluating the CEO, we enter both types of ROA variables in the same equation. The two ROA variables are orthogonalized using Stata’s orthog command before the estimation. As the main effect of ROA relative to aspiration levels is negative (i.e., a lower ROA relative to aspiration levels increases the risk of CEO dismissal), the interaction effect with demographic faultlines is expected to have a positive sign, meaning faultlines weaken the effect of low performance on CEO dismissal. The interaction of ROA relative to social aspiration levels and demographic faultline strength is negative and not significant, whereas the coefficient for the interaction of ROA relative to historical aspiration levels and demographic faultline strength is positive and significant (β = 1.92, p = .001). This is partially consistent with Hypothesis 1. Model 4 includes a variable for information-related faultline strength, which is negative but not statistically significant. Model 5 tests Hypothesis 2, which predicted that the effect of firm performance below the aspiration level is amplified when board information-related faultlines are stronger. As a partial support for Hypothesis 2, the coefficient for the interaction of ROA relative to social aspiration levels and information-related faultline strength is negative and significant (β = −2.99, p = .000), whereas the interaction of ROA relative to historical aspiration and information-related faultlines is positive and not significant. Model 6 presents a fully saturated model with both demographic and information-related faultlines as well as their interaction terms with ROA relative to social and historical aspiration levels. Similar to Models 3 and 5, the interaction of ROA relative to historical aspiration and demographic faultline is positive and significant (β = 1.83, p = .000), and the interaction of ROA relative to social aspiration and information-related faultline is negative and significant (β = −2.75, p = .000).

OPEN IN VIEWER

Table 2

Cox Regression Results Predicting Hazard Rate of CEO Dismissal

Variable	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Log of total firm salesp value	.24 (.10).014	.24 (.10).015	.24 (.10).019	.25 (.10).013	.23 (.10).020	.23 (.10).024
CEO agep value	–.01 (.02).672	–.01 (.02).660	–.01 (.02).635	–.01 (.02).673	–.01 (.02).625	–.01 (.02).586
Female CEOp value	.35 (.66).596	.35 (.66).591	.51 (.65).438	.32 (.67).631	.29 (.67).671	.42 (.66).520
Outside CEOp value	.45 (.20).027	.45 (.20).027	.44 (.20).030	.45 (.20).028	.46 (.20).026	.44 (.20).030
CEO–chair dualityp value	–.20 (.20).312	–.20 (.20).309	–.23 (.20).257	–.20 (.20).309	–.22 (.20).277	–.23 (.20).258
CEO ownershipp value	–.14 (.09).102	–.14 (.09).103	–.14 (.09).115	–.14 (.08).102	–.13 (.08).101	–.14 (.08).109
Average director ownershipp value	–.02 (.07).780	–.02 (.07).744	–.03 (.08).705	–.02 (.07).793	.00 (.06).981	–.00 (.06).986
Average director tenurep value	–.05 (.04).193	–.05 (.04).185	–.05 (.04).182	–.05 (.04).200	–.05 (.04).179	–.05 (.04).199
Proportion directors after CEOp value	–.36 (.43).404	–.36 (.44).409	–.39 (.44).372	–.34 (.44).447	–.32 (.45).473	–.30 (.44).507
Average director external directorshipsp value	–.27 (.21).202	–.25 (.21).228	–.20 (.21).328	–.26 (.21).215	–.26 (.21).212	–.19 (.21).359
Board sizep value	–.01 (.04).909	–.00 (.05).935	–.00 (.04).983	–.00 (.04).953	.00 (.04).957	.00 (.04).946
CEO–board similarityp value	.77 (.35).030	.77 (.35).029	.87 (.38).022	.77 (.35).029	.74 (.37).046	.84 (.38).029
Director age heterogeneityp value	−1.72 (3.30).601	−1.50 (3.35).654	−1.15 (3.48).741	−1.66 (3.32).618	−1.14 (3.33).731	–0.67 (3.44).845
Proportion female directorsp value	4.36 (1.55).005	4.32 (1.55).005	4.61 (1.69).006	4.40 (1.56).005	4.43 (1.64).007	4.71 (1.72).006
Director racial heterogeneityp value	.28 (.94).769	.25 (.94).787	.36 (.96).706	.27 (.95).778	.20 (.97).837	.31 (.99).756
Proportion independent directorsp value	–.04 (.76).954	–.07 (.75).926	–.01 (.77).985	–.07 (.76).929	–.07 (.75).924	–.05 (.76).944
Director job title heterogeneityp value	.44 (.62).477	.46 (.61).457	.61 (.61).315	.44 (.62).474	.44 (.61).468	.63 (.61).300
Director tenure heterogeneityp value	.99 (.45).026	1.00 (0.45).026	1.04 (0.45).020	1.06 (0.47).024	1.08 (0.46).019	1.09 (0.46).018
Shareholder returns (social aspiration)p value	–.55 (.11).000	–.55 (.11).000	–.54 (.11).000	–.55 (.11).000	–.56 (.11).000	–.55 (.11).000
Shareholder returns (historical aspiration)p value	–.18 (.10).086	–.18 (.10).083	–.18 (.10).074	–.18 (.10).086	–.18 (.10).075	–.19 (.10).061
ROA (social aspiration)p value	–.31 (.09).000	–.31 (.09).000	–.29 (.08).000	–.31 (.09).000	–.32 (.08).000	–.33 (.08).000
ROA (historical aspiration)p value	–.41 (.06).000	–.41 (.06).000	–.44 (.06).000	–.41 (.06).000	–.43 (.05).000	–.46 (.05).000
Demographic fault linep value		–0.66 (1.16).570	.40 (.98).682			.95 (.99).338
ROA (Social Aspiration) × Demographic Fault Linep value			–.81 (.82).327			.64 (.96).503
ROA (Historical Aspiration) × Demographic Fault Linep value			1.92 (0.60).001			1.83 (0.49).000
Information-related fault linep value				–0.60 (1.12).592	–0.75 (1.04).470	–0.43 (1.09).692
ROA (Social Aspiration) × Information-Related Fault Linep value					−2.99 (0.70).000	−2.75 (0.69).000
ROA (Historical Aspiration) × Information-Related Fault Linep value					.33 (.41).421	.61 (.41).139
Number of firm-year observations	5,951	5,951	5,951	5,949	5,949	5,949
Number of CEOs	1,057	1,057	1,057	1,057	1,057	1,057
Number of events	144	144	144	144	144	144
Number of months in risk set	67,255	67,255	67,255	67,231	67,231	67,231
Log likelihood	−786.07	−785.90	−782.08	−785.87	−780.30	−776.18

Note: Standard errors are in parentheses. All models include year dummies and two-digit Standard Industrial Classification industry dummies. All explanatory variables are lagged by 1 year. ROA = return on assets.

Figures 1 and 2 illustrate the moderating effects of faultlines. Based on the results of Model 3 in Table 2, the graphs present estimated cumulative hazard function by demographic faultline strength and firm performance relative to historical aspiration level. The cumulative hazard of being dismissed increased as the CEO remained on the job, but the rate of dismissal varied by demographic faultlines and firm performance. Figure 1 shows the moderating effect of demographic faultlines. Comparing the two groups where demographic faultlines were weak (i.e., one standard deviation below the mean), the effect of ROA relative to aspirations was large; the board’s decision to dismiss the CEO was sensitive to low firm performance. In contrast, between the two groups where demographic faultlines were strong (i.e., one standard deviation above the mean), the effect of ROA relative to aspirations was small, suggesting that strong faultlines undermined the board’s ability to take a decisive action. Figure 2 shows the moderating effect of information-based faultlines. Between the two groups with weak information-related faultlines (i.e., one standard deviation below the mean), the effect of ROA relative to social aspiration levels was small. When information-related faultlines were strong (i.e., one standard deviation above the mean), ROA relative to aspiration levels had a greater impact on CEO dismissal. Had we not distinguished demographic and information-related faultlines, we would have confounded the distinct effects of the two types and concluded that overall faultlines do not affect CEO dismissal.

OPEN IN VIEWER

Figure 1

Estimated Cumulative Hazard Function for Strong Versus Weak Demographic Faultlines by Return on Assets Relative to Historical Aspiration Levels

OPEN IN VIEWER

Figure 2

Estimated Cumulative Hazard Function for Strong Versus Weak Information-Related Faultlines by Return on Assets Relative to Social Aspiration Levels

As a robustness check, we also tested whether there are moderating effects of traditional diversity variables by entering interaction terms between ROA relative to social and historical aspiration levels and each of the six traditional measures of diversity (i.e., age heterogeneity, proportion of female directors, race heterogeneity, proportion of independent outsiders, job title heterogeneity, and tenure heterogeneity). None of these interaction variables was statistically significant. Overall, the results suggest that faultline measures reveal group dynamics that are distinct from traditional diversity measures.

If board faultlines actually had an impact on CEO dismissal, then faultlines should not affect other types of CEO departures or they should have different effects on other types of departures. If the faultline effect that we report was indeed spurious, then faultlines might also affect nondismissal departures. As a placebo test, similar to Hubbard et al.’s (2017) analysis, we estimated Cox models predicting nondismissal CEO departures, including departures due to retirement and voluntary turnover. Table 3 presents the results. In all models of Table 3, the results are remarkably different from those of CEO dismissal models. As expected, older CEOs have higher rates of nondismissal departures. None of the interactions between ROA and faultlines are significant at the 5% level. These results add confidence to our findings that board faultlines have significant moderating effects on CEO dismissal but not on other types of CEO turnover.

OPEN IN VIEWER

Table 3

Cox Regression Results Predicting Hazard Rate of Nondismissal Turnover

Variable	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Log of total firm salesp value	–.04 (.06).483	–.04 (.06).485	–.04 (.06).478	–.04 (.06).497	–.04 (.06).498	–.04 (.06).495
CEO agep value	.14 (.01).000	.14 (.01).000	.14 (.01).000	.14 (.01).000	.14 (.01).000	.14 (.01).000
Female CEOp value	.31 (.46).494	.31 (.46).491	.32 (.46).488	.31 (.46).497	.32 (.46).487	.33 (.46).480
Outside CEOp value	.02 (.14).900	.02 (.14).894	.02 (.14).889	.02 (.14).885	.02 (.14).904	.02 (.14).891
CEO–chair dualityp value	–.08 (.12).495	–.08 (.12).496	–.09 (.12).479	–.09 (.12).482	–.09 (.12).446	–.09 (.12).429
CEO ownershipp value	–.04 (.02).068	–.04 (.02).069	–.04 (.02).072	–.04 (.02).069	–.04 (.02).065	–.04 (.02).070
Average director ownershipp value	.04 (.02).054	.04 (.02).053	.04 (.02).054	.04 (.02).053	.04 (.02).051	.04 (.02).052
Average director tenurep value	–.07 (.02).003	–.07 (.02).003	–.07 (.02).003	–.06 (.02).003	–.06 (.02).004	–.06 (.02).005
Proportion directors after CEOp value	.10 (.24).682	.10 (.24).683	.10 (.24).680	.13 (.25).592	.14 (.25).583	.14 (.25).580
Average director external directorshipsp value	.25 (.10).016	.25 (.10).017	.25 (.10).016	.25 (.10).015	.26 (.10).012	.26 (.10).012
Board sizep value	.02 (.03).480	.02 (.03).489	.02 (.03).470	.02 (.03).453	.02 (.03).430	.02 (.03).422
CEO–board similarityp value	.23 (.21).272	.23 (.21).276	.23 (.21).280	.24 (.21).254	.27 (.20).191	.26 (.20).197
Director age heterogeneityp value	−2.46 (1.92).199	−2.53 (1.93).189	−2.56 (1.92).184	−2.36 (1.93).222	−2.59 (1.93).180	−2.68 (1.95).169
Proportion female directorsp value	2.36 (0.93).012	2.36 (0.94).012	2.36 (0.93).011	2.37 (0.94).011	2.47 (0.91).007	2.47 (0.92).007
Director racial heterogeneityp value	–.18 (.48).701	–.18 (.48).717	–.18 (.49).706	–.18 (.48).711	–.15 (.48).759	–.15 (.48).761
Proportion independent directorsp value	–.43 (.45).346	–.42 (.45).353	–.42 (.45).350	–.42 (.45).352	–.39 (.45).381	–.39 (.45).392
Director job title heterogeneityp value	–.43 (.35).216	–.43 (.35).214	–.43 (.35).219	–.45 (.35).199	–.48 (.35).176	–.48 (.35).178
Director tenure heterogeneityp value	.08 (.24).735	.08 (.24).741	.08 (.24).738	.11 (.25).651	.09 (.25).710	.09 (.25).713
Shareholder returns (social aspiration)p value	.08 (.04).045	.08 (.04).046	.08 (.04).042	.08 (.04).046	.08 (.04).046	.08 (.04).043
Shareholder returns (historical aspiration)p value	–.08 (.06).175	–.08 (.06).172	–.08 (.06).172	–.08 (.06).171	–.07 (.06).186	–.07 (.06).184
ROA (social aspiration)p value	–.08 (.06).163	–.08 (.06).169	–.08 (.07).235	–.08 (.06).160	–.07 (.05).185	–.06 (.05).259
ROA (historical aspiration)p value	–.06 (.05).258	–.06 (.05).261	–.06 (.05).253	–.06 (.05).254	–.06 (.05).245	–.06 (.05).234
Demographic fault linep value		.19 (.64).765	.20 (.63).753			.21 (.63).744
ROA (Social Aspiration) × Demographic Fault Linep value			–.22 (.81).787			–.21 (.57).717
ROA (Historical Aspiration) × Demographic Fault Linep value			.33 (.72).648			.35 (.61).565
Information-related fault linep value				–.42 (.60).479	–.39 (.60).521	–.39 (.60).516
ROA (Social Aspiration) × Information-Related Fault Linep value					.78 (.44).077	.78 (.43).072
ROA (Historical Aspiration) × Information-Related Fault Linep value					.33 (.35).348	.34 (.34).324
Number of firm-year observations	5,951	5,951	5,951	5,949	5,949	5,949
Number of CEOs	1,057	1,057	1,057	1,057	1,057	1,057
Number of events	433	433	433	433	433	433
Number of months in risk set	67,255	67,255	67,255	67,231	67,231	67,231
Log likelihood	−2,379.07	−2,379.03	−2,378.80	−2,378.77	−2,376.71	−2,376.42

Note: Standard errors are in parentheses. All models include year dummies and two-digit Standard Industrial Classification industry dummies. All explanatory variables are lagged by 1 year. ROA = return on assets.

As an additional robustness check, we converted the continuous-time duration data into annual panel data and estimated discrete-time event history analysis. As Allison (1982) noted, a discrete-time event history analysis can approximate a continuous-time analysis and produce estimates that are consistent and asymptotically efficient. In a discrete-time version, the hazard rate is the conditional probability that a CEO is dismissed in the year, given that the CEO has not already been dismissed. The time variable is the CEO’s tenure at the firm, measured in years. Table 4 presents the results of complementary log-log regression models predicting the probability of CEO dismissal. The models include CEO tenure and tenure-squared variables to allow baseline hazards to change over the CEO’s tenure. The results in Table 4 are consistent with the continuous-time Cox regression results in Table 2, supporting Hypotheses 1 when historical aspiration levels are used (Model 3) and supporting Hypothesis 2 when social aspiration levels are used (Model 5). A fully saturated model (Model 6) also supports the hypotheses. As a robustness check, we also estimated logit regression and random-effects logit models (Wiersema & Zhang, 2011), which produced similar results (see Appendix Tables A4 and A5). ⁶

OPEN IN VIEWER

Table 4

Complementary Log-Log Regression Results Predicting CEO Dismissal

Variable	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Log of total firm salesp value	.24 (.10).014	.23 (.10).016	.23 (.10).020	.25 (.10).012	.23 (.10).018	.23 (.10).022
CEO agep value	–.04 (.02).042	–.04 (.02).039	–.05 (.02).035	–.04 (.02).042	–.04 (.02).038	–.05 (.02).030
Female CEOp value	.24 (.60).684	.26 (.61).666	.38 (.59).519	.19 (.62).759	.15 (.62).805	.24 (.59).683
Outside CEOp value	.41 (.21).055	.40 (.21).056	.39 (.21).063	.40 (.21).058	.40 (.21).062	.39 (.21).067
CEO–chair dualityp value	–.20 (.20).325	–.19 (.21).344	–.21 (.21).306	–.21 (.21).310	–.22 (.20).278	–.21 (.21).305
CEO ownershipp value	–.17 (.10).073	–.17 (.10).073	–.18 (.10).076	–.16 (.09).073	–.15 (.09).083	–.16 (.09).085
Average director ownershipp value	–.02 (.06).799	–.02 (.06).755	–.03 (.07).678	–.01 (.06).815	.00 (.05).994	–.00 (.05).968
Average director tenurep value	–.06 (.04).099	–.07 (.04).094	–.06 (.04).151	–.06 (.04).102	–.06 (.04).091	–.06 (.04).147
Proportion directors after CEOp value	–.61 (.44).168	–.60 (.44).174	–.59 (.45).190	–.56 (.45).208	–.52 (.44).238	–.46 (.44).304
Average director external directorshipsp value	–.25 (.21).235	–.23 (.21).266	–.18 (.21).375	–.24 (.21).256	–.24 (.21).250	–.19 (.21).370
Board sizep value	–.02 (.05).719	–.01 (.05).765	–.01 (.04).775	–.01 (.04).770	–.01 (.04).844	–.01 (.04).861
CEO–board similarityp value	.84 (.33).011	.86 (.33).010	1.04 (0.36).004	.84 (.33).011	.81 (.35).021	.91 (.35).009
Director age heterogeneityp value	−2.98 (3.29).365	−2.62 (3.31).429	−2.41 (3.45).485	−2.99 (3.30).365	−2.27 (3.32).494	−2.16 (3.43).529
Proportion female directorsp value	4.10 (1.40).003	4.02 (1.41).004	4.53 (1.53).003	4.17 (1.42).003	4.11 (1.48).005	4.26 (1.52).005
Director racial heterogeneityp value	.48 (.86).578	.45 (.85).593	.76 (.90).395	.46 (.87).601	.42 (.90).640	.61 (.92).508
Proportion independent directorsp value	.12 (.77).879	.08 (.77).914	.18 (.78).821	.08 (.77).917	.08 (.77).922	.08 (.78).915
Director job title heterogeneityp value	.59 (.63).350	.62 (.62).320	.71 (.61).247	.58 (.63).354	.51 (.63).416	.68 (.62).272
Director tenure heterogeneityp value	1.12 (0.44).011	1.12 (0.44).011	1.16 (0.44).008	1.22 (0.46).008	1.22 (0.45).007	1.20 (0.45).008
CEO tenurep value	.08 (.05).116	.08 (.05).113	.08 (.05).106	.08 (.05).127	.08 (.05).142	.08 (.05).125
CEO Tenure × CEO Tenurep value	–.00 (.00).635	–.00 (.00).638	–.00 (.00).639	–.00 (.00).663	–.00 (.00).661	–.00 (.00).646
Shareholder returns (social aspiration)p value	–.59 (.12).000	–.59 (.12).000	–.61 (.12).000	–.59 (.12).000	–.60 (.12).000	–.6 (.13).000
Shareholder returns (historical aspiration)p value	–.24 (.10).017	–.24 (.10).017	–.24 (.10).018	–.24 (.10).017	–.27 (.10).010	–.26 (.10).010
ROA (social aspiration)p value	–.27 (.08).001	–.27 (.08).001	–.25 (.07).001	–.27 (.08).001	–.27 (.07).000	–.28 (.07).000
ROA (historical aspiration)p value	–.39 (.05).000	–.39 (.06).000	–.42 (.06).000	–.39 (.05).000	–.40 (.05).000	–.44 (.05).000
Demographic fault linep value		−1.13 (1.20).349	0.12 (1.04).905			.46 (.99).640
ROA (Social Aspiration) × Demographic Fault Linep value			–.50 (.63).427			.75 (.82).359
ROA (Historical Aspiration) × Demographic Fault Linep value			2.05 (0.58).000			2.19 (0.55).000
Information-related fault linep value				−1.03 (1.13).364	−1.20 (1.08).267	–0.87 (1.05).411
ROA (Social Aspiration) × Information-Related Fault Linep value					−2.86 (.062).000	−2.75 (0.57).000
ROA (Historical Aspiration) × Information-Related Fault Linep value					.24 (.38).523	.43 (.34).207
Constantp value	−2.94 (1.55).058	−2.92 (1.55).060	−3.16 (1.59).047	−3.11 (1.56).046	−3.10 (1.53).043	−3.18 (1.57).043
Number of firm-year observations	4,804	4,804	4,804	4,800	4,800	4,800
Number of events	144	144	144	144	144	144
Number of CEOs	913	913	913	913	913	913
Log likelihood	−531.17	−530.68	−525.98	−530.62	−525.39	−519.69

Note: Standard errors are in parentheses. All models include year dummies and two-digit Standard Industrial Classification industry dummies. All explanatory variables are lagged by 1 year. ROA = return on assets.

Using the complementary log-log regression results, Figure 3 shows estimated dismissal probabilities as a function of firm performance. We used Stata’s margins and marginsplot commands. The graph shows that the effect of ROA relative to historical aspiration levels on CEO dismissal is generally negative, but it varies by demographic faultline strength. When demographic faultlines are weak (i.e., below mean), dismissal probability increases as ROA becomes lower than the aspiration level. However, when the faultlines are strong (i.e., above mean), low ROA relative to the aspiration level does not necessarily lead to CEO dismissal. Strong demographic faultlines dampen the effect of low performance, which is consistent with Hypothesis 1. At the lowest level of ROA relative to the aspiration level (i.e., −4), strong demographic faultlines can reduce dismissal probability by 8.5 percentage points (.169 – .084 = .085), a 50% decrease in dismissal risk. Figure 4 illustrates the moderating effect of information-related faultlines. When information-related faultlines become strong (i.e., above mean), the negative relationship between ROA relative to the aspiration level and dismissal probabilities becomes stronger. Information-related faultlines magnify the sensitivity of CEO dismissal to poor firm performance, which supports Hypothesis 2. At the lowest level of ROA relative to the aspiration level (i.e., −4), strong information-related faultlines can increase dismissal probability by 9.2 percentage points (.133 – .041 = .092), a 69% increase in dismissal risks. When evaluated at a less extreme value of ROA (i.e., −2), strong demographic faultlines decrease dismissal probability by 1.6 percentage points (.067 – .051 = .016, a 24% decrease), and strong information-related faultlines increase dismissal probability by 2.5 percentage points (.061 – .036 = .025, a 41% increase). While the changes in dismissal probabilities may seem a modest magnitude, given that CEO dismissal is a relatively infrequent event, with only 3% of firm-year observations being dismissal cases, this is a substantively meaningful effect.

OPEN IN VIEWER

Figure 3

CEO Dismissal Probabilities as a Function of Return on Assets Relative to Historical Aspiration Levels Separately by Demographic Faultline Strength

OPEN IN VIEWER

Figure 4

CEO Dismissal Probabilities as a Function of Return on Assets Relative to Social Aspiration Levels Separately by Information-Related Faultline Strength

To evaluate the robustness of causal inferences, we assessed what proportion of an estimate must be due to bias to invalidate the inference (Busenbark, Yoon, Gamache, & Withers, 2022; Frank, Maroulis, Duong, & Kelcey, 2013). Based on Rubin’s causal model (Rubin, 1974), this approach asks how one could invalidate inferences by replacing observed cases with unobserved cases in which there was no causal effect. We used Xu, Frank, Maroulis, and Rosenberg’s (2019) Stata command confound, which performs Frank et al.’s (2013) algorithm to quantify the proportion of observed cases that would have to be replaced with null hypothesis cases to invalidate the inference. ⁷ The results suggest that to invalidate the inference about the moderating effect of demographic faultlines, 41.6% of the cases (or 1,998 firm-years in the complementary log-log model) would have to be replaced with cases for which there is an effect of zero. To invalidate the inference about the moderating effect of information-related faultlines, 47.2% of the cases (or 2,267 firm-years in the complementary log-log model) would have to be replaced with cases for which there are no effects. Intuitively speaking, to believe that the interaction of faultlines and firm performance did not have any effect on CEO dismissal, one must accept that outcomes for almost half of the cases in the sample were generated by some other causes that were independent of board faultlines and firm performance. This observation strengthens our confidence in the causal inference.

Finally, endogeneity of our main independent and moderator variables may bias the results. As boards are not randomly assigned to strong or weak faultlines, this situation can be seen as selection of treatment. This is also considered as a case of omitted variables, as unobservable factors that determine nonrandom assignment of treatment may have not been included in the model. If such omitted variables affect firm performance and are correlated with board faultlines, the observed predictor—board faultlines—is correlated with the unobserved residual, which causes bias in the coefficient estimates (Hill, Johnson, Greco, O’Boyle, & Walter, 2021).

As a remedy for endogeneity arising from selection of treatment and omitted variables, we used instrumental-variable regression approach. We identified four instrumental variables that can affect board faultlines: board size, director turnover, director appointments, and directors’ external board membership. Since group size is a fundamental attribute that determines relationships among individuals and subgroups, board size—measured as the number of directors—is expected to be correlated with board faultline strengths (Ali & Ayoko, 2020). Director turnover and appointments directly alter board composition, thereby affecting board faultlines. Director turnover is defined as the number of directors who left the board in the previous year. Director appointment is measured as the proportion of directors appointed after the CEO took office. Finally, directors with board directorships at other firms may have social influence over the other directors as well as difficulties in balancing multiple commitments at different boards (Cashman, Gillan, & Jun, 2012; He & Huang, 2011), which may affect interpersonal relationships and faultlines within the focal board. Director external directorships are measured as the average number of director positions outside the focal firm held by outside directors. For the instrumental-variable regression models treating the interaction of ROA and demographic faultlines as endogenous, we used all four instruments. For the models treating the interaction of ROA and information-related faultlines as endogenous, we used board size, director turnover, and director appointments as the instruments. We chose the instruments in each model based on the relevance and exogeneity of the instruments (Semadeni, Withers, & Certo, 2014). Each of the instruments is significantly related to the endogenous predictors (i.e., faultlines and the interactions of firm performance and faultlines). A Wald test of exogeneity of the instruments rejects the null hypothesis of no endogeneity (p < .01; see Table 5). Besides, it is reasonable to argue that the instruments—board size, director turnover, director appointments, and external directorships—are not correlated with the residual for CEO dismissal (Dalton, Daily, Ellstrand, & Johnson, 1998; Dalton, Daily, Johnson, & Ellstrand, 1999).

OPEN IN VIEWER

Table 5

Instrumental Variable Probit Regression Results Predicting CEO Dismissal

Variable	Model 1	Model 2	Model 3	Model 4
Log of total firm salesp value	.03 (.02).094	–.01 (.02).577	.01 (.02).579	–.00 (.02).812
CEO agep value	.00 (.00).723	.00 (.00).227	–.01 (.00).015	–.01 (.00).064
Female CEOp value	.13 (.14).380	.27 (.13).043	–.06 (.14).670	–.11 (.14).415
Outside CEOp value	–.02 (.05).808	.03 (.04).487	–.01 (.04).758	.06 (.04).170
CEO–chair dualityp value	–.03 (.04).534	–.08 (.04).024	.00 (.04).954	.03 (.04).391
CEO ownershipp value	–.03 (.01).009	.01 (.01).489	–.00 (.01).738	–.01 (.01).194
Average director ownershipp value	–.00 (.01).903	–.00 (.01).821	.01 (.01).376	.01 (.01).153
Average director tenurep value	–.01 (.01).434	–.00 (.01).965	.00 (.01).608	.00 (.01).911
CEO–board similarityp value	.03 (.07).708	.09 (.06).119	–.00 (.06).966	–.04 (.06).523
Director age heterogeneityp value	.79 (.90).378	−1.49 (0.59).012	1.91 (0.68).005	.29 (.65).655
Proportion female directorsp value	–.06 (.30).851	.53 (.27).053	.09 (.30).751	–.19 (.28).511
Director racial heterogeneityp value	–.17 (.17).314	.23 (.15).117	–.22 (.16).154	.05 (.15).724
Proportion independent directorsp value	.09 (.16).578	.26 (.14).056	–.11 (.15).477	–.16 (.14).262
Director job title heterogeneityp value	.17 (.11).138	.12 (.11).258	–.02 (.11).891	.03 (.11).804
Director tenure heterogeneityp value	.08 (.09).392	.05 (.08).579	.12 (.12).315	.03 (.11).776
CEO tenurep value	.01 (.01).251	.01 (.01).067	–.01 (.01).465	.01 (.01).369
CEO Tenure × CEO Tenurep value	.00 (.00).858	–.00 (.00).197	.00 (.00).751	–.00 (.00).344
Shareholder returns (social aspiration)p value	–.02 (.03).376	–.03 (.02).252	–.01 (.03).798	–.03 (.03).236
Shareholder returns (historical aspiration)p value	.01 (.02).529	–.09 (.02).000	–.07 (.02).002	.04 (.02).044
ROA (social aspiration)p value	–.15 (.02).000	.02 (.02).230	–.25 (.02).000	–.03 (.02).046
ROA (historical aspiration)p value	–.02 (.02).458	–.11 (.02).000	–.05 (.02).025	–.09 (.02).000
Demographic fault linep value	−4.34 (1.87).020	3.02 (0.47).000
ROA (Social Aspiration) × DemographicFault Linep value	9.76 (0.44).000
ROA (Historical Aspiration) × DemographicFault Linep value		10.73 (0.15).000
Average director external directorshipsp value			–.05 (.04).147	–.05 (.04).192
Information-related fault linep value			–0.68 (1.24).585	0.02 (1.07).985
ROA (Social Aspiration) × Information-Related Fault Linep value			−8.41 (0.13).000
ROA (Historical Aspiration) × Information-Related Fault Linep value				−8.65 (0.10).000
Constantp value	–.60 (.35).085	–.28 (.33).406	.17 (.42).675	.41 (.38).276
Number of firm-year observations	4,417	4,417	4,415	4,415
Log likelihood	8,600.30	8,961.08	7,097.83	7,247.52
Wald test χ2	895.67	1,091.06	48.81	13.11

Note: Standard errors in parentheses. All models include year dummies and two-digit Standard Industrial Classification industry dummies. All explanatory variables are lagged by 1 year. ROA = return on assets.

As the dependent variable in the second stage is CEO dismissal, a binary variable, we estimated probit models with endogenous independent variables using Stata’s ivprobit command. Table 5 presents the results using maximum-likelihood estimation. Models 1 and 2 treat demographic faultlines and the interaction of ROA and demographic faultlines as endogenous, using board size, director turnover, director appointments, and external directorships as instruments. The results indicate that the interaction of ROA and demographic faultlines is positive and significant, supporting Hypothesis 1. Models 3 and 4 treat information-related faultlines and its interaction with ROA as endogenous, using board size, director turnover, and director appointments as instruments. The results show that the interaction of ROA and information-related faultlines is negative and significant, consistent with Hypothesis 2. ⁸ Weak-instrument robust tests, such as the Anderson-Rubin test, overidentification test, and the conditional likelihood-ratio test, reject the null hypothesis that the coefficients for the endogenous independent variables are jointly zero, using Stata’s weakiv command (Finlay, Magnusson, & Schaffer, 2013). While we acknowledge that we are unable to eliminate all sources of endogeneity and that the validity of instrumental variables regression depends on meeting the exogeneity condition and ensuring instrument strengths (Hill et al., 2021; Semadeni et al., 2014), the results of the instrumental variables regression add confidence to our argument regarding the effects of board faultlines.

Discussion and Conclusion

Group-level processes have attracted much attention from management researchers. Interpersonal relations within groups can significantly affect processes within the group, altering outcomes such as group decision-making (Forbes & Milliken, 1999; Haleblian & Rajagopalan, 2006). In particular, faultlines based on multiple characteristics of the directors pose a promising novel concept that highlights important dynamics within the board, especially in the context of making critical decisions, such as CEO dismissal.

In this study, we examined how board faultlines affect decisions about CEO dismissal. Drawing on social identity theory and self-categorization theory, we predicted that strong demographic faultlines in a board can hamper cooperation and increase conflict, undermining a board’s ability to make decisive action to dismiss the CEO when the financial performance of the firm is below the aspiration level. Based on the CEM, we also hypothesized that information-based faultlines can enhance the benefit of exchanging diverse information and expertise in a board, enhancing board ability to evaluate wider information and reach a higher-quality decision regarding CEO dismissal. Using data from S&P 500 boards, we found that demographic faultlines of boards attenuated the relationship between firm performance below historical aspiration levels and CEO dismissal, whereas information-related faultlines strengthened the effect of firm performance below social aspiration levels on CEO dismissal. The two types of faultlines (i.e., demographic and information-related faultlines) had distinct effects in opposite directions, which have not been disentangled in prior studies.

Our findings highlight board faultlines as a significant moderator of the relationship between firm performance and CEO dismissal. The literature on CEO dismissal suggests a need to study boundary conditions and contexts that moderate the relationship between firm performance and CEO dismissal (Brickley, 2003; Fredrickson et al., 1988; Hilger et al., 2013). Although the research has identified several important moderators at the firm level, there has been little work on group dynamics within boards. By discovering the role of board faultlines as an important moderator, we contribute to scholarly efforts in strategy and corporate governance to understand group- and team-level board dynamics (Forbes & Milliken, 1999; Payne, Benson, & Finegold, 2009), particularly, cognitive aspects of CEO dismissal (Graffin et al., 2013; Haleblian & Rajagopalan, 2006; Zhang, 2008).

We argue that the faultlines perspective offers important advantages that can overcome some long-standing limitations in the research about corporate upper echelons, such as TMTs and boards. First, as a group-level concept, faultlines depict structural attributes of the group that cannot be reduced to individual members. This is distinct from the aggregation of individual member attributes using measures such as the coefficient of variation or the index of heterogeneity, an approach that has been popular in the literature. By lifting the level of analysis to the group level, the faultlines approach presents an alternative path to opening up the “black box” of upper-echelon teams (Hambrick, 2007). Rather than studying microlevel traits, such as cognition and personality, the faultline approach enables researchers to study group-level characteristics, including the configuration of subgroups, intersubgroup relationships, and the activation of faultlines within the group. Second, the faultlines approach enables an examination of multiple attributes of individual members simultaneously. Existing research on diversity and heterogeneity in groups tends to focus on specific attributes, such as race, gender, or tenure. However, individuals rarely perceive others along a single attribute; the identity of each individual is constructed along multiple dimensions (Hillman et al., 2008). Examining multiple attributes simultaneously as a set opens up an exciting new avenue of research leading to a realistic picture of how groups function.

We found that demographic and information-related faultlines moderate the relationship between firm performance and CEO dismissal in opposite ways. This highlights the importance of separating the two types of faultlines. An analysis that examines only one type of faultlines, or an analysis that confounds the two types of faultlines by combining them under one category, would be incomplete or misleading. This may explain the mixed findings in the existing literature (Thatcher & Patel, 2012), in which some studies reported positive effects of faultlines while others found negative effects. Our post hoc interpretation is that our findings reflect the fundamental difference between separation-based and variety-based faultlines (Carton & Cummings, 2012). Building on Harrison and Klein’s (2007) typology of diversity—separation, disparity, and variety—Carton and Cummings (2012) argued that different types of faultlines have different causal effects on intergroup processes. Our demographic faultline measure corresponds to separation-based faultlines that are based on horizontal differences among board members. Our information-related faultline measure represents variety-based faultlines that are based on qualitatively distinct knowledge possessed by board members.

The perspective that views demographic faultlines as based on separation and information-based faultlines as variety may provide a post hoc explanation regarding why the moderating effect of faultlines is significant only for a certain type of firm performance measures. Specifically, demographic faultlines have a significant moderating effect only with performance relative to historical aspirational levels, whereas information-related faultlines have a significant moderating effect only with performance relative to social aspirational levels.

Compared with social aspiration levels, historical aspiration levels are more readily available and easily interpretable internal information that directors use when evaluating a CEO without much knowledge about external environments. Because historical aspiration levels do not require complex elaboration process, when directors rely on historical track record to evaluate the CEO, the board may naturally be divided along separation-based faultlines (i.e., demographic faultlines) based on salient demographic characteristics. The faultlines fragment the board’s collective identity and heighten the perceived threat to each subgroup’s identity, thereby making critical decision-making difficult.

Social aspirations, which require external information, can be better understood and utilized as a way to evaluate the CEO in the board where both internal and external task-related information is actively shared and elaborated. Thus, when directors rely on social aspiration levels to evaluate the CEO, the board may be fragmented along variety-based faultlines (i.e., information-related faultlines), which may expose the directors to new perspectives and facilitate knowledge exchange among directors. As variety of information and perspectives often originates from sources outside the organization, the beneficial effect of variety-based faultlines might have been more salient when directors evaluate the CEO relative to peer CEOs, rather than when they consider the historical track record. With more refined measures, future research should validate our interpretation. Direct observations of board deliberation process, for example, board meeting transcripts, may allow researchers to examine the role of historical versus social aspirations. Such explorations can contribute to our understanding of performance aspirations and group faultlines in the context of board decision-making.

Based on our findings about demographic faultlines, we argue that cliques within boards can undermine cohesion and increase conflicts within the board, which in turn can weaken a board’s ability to execute a plan to dismiss the CEO. Because a failure to dismiss the CEO when necessary can be detrimental to the firm, boards afflicted by demographic faultlines should pay extra attention to the coordination process when they make important decisions, such as CEO dismissal. A fundamental solution is to form and maintain a board in a way that does not develop strong cliques whose intersubgroup differences are highly salient. According to Lau and Murnighan (1998: 331), faultlines are most likely in groups of moderate diversity and absent or unlikely in situations of minimum and maximum diversity. Considering that faultlines can be both strong and weak at the same level of diversity, the negative effect of demographic faultlines may be mitigated by a board’s willingness and efforts to prevent faultlines from becoming stronger. This offers a practical implication for corporate boards; when selecting new directors, it would be more useful to consider potential faultlines in the board as a whole rather than to focus on individual directors’ demographic attributes.

On the other hand, we found that information-based faultlines within boards can enhance a board’s ability to exchange diverse information and perspectives, facilitating a timely decision to dismiss a CEO. This is consistent with the literature showing that information-based diversity and faultlines are positively related to group performance when the task is complex and when creativity and innovation are important dimensions of performance (Gibson & Vermeulen, 2003; Jehn et al., 1999), which are the conditions that characterize board decision-making regarding CEO evaluation. Our findings highlight the importance of nondemographic attributes in boardrooms. In recent years, management scholars, practitioners, and legislators have shown great interest in increasing demographic diversity in the corporate upper echelons by bringing more women and ethnic minorities to top positions (Cutter, 2020; Hoobler, Masterson, Nkomo, & Michel, 2018; Knippen et al., 2019). Compared with the intensive focus on demographic attributes, relatively little attention has been paid to diversity based on nondemographic attributes, such as work experience, education, and primary job title. ⁹ Our results imply that faultlines can be a double-edged sword with both positive and negative potentials, which is analogous to what diversity research has demonstrated (Horwitz & Horwitz, 2007; Webber & Donahue, 2001). When demographic diversity allows the team to split into factional subgroups, the negative effect may become predominant. Information-related diversity, in contrast, is less likely to create harmful subgroups and still has beneficial effects on team performance by bringing in novel ideas and facilitating innovation. Our findings about the positive effect of information-based faultlines respond to the calls to “develop and test models of faultlines that examine potential positive effects of faultlines.” (Thatcher & Patel, 2012: 999) Based on our findings, we argue that TMTs and boards need to foster greater nondemographic diversity and carefully monitor development of factional subgroups that are based solely on demographic attributes.

We should note our study’s limitations. First, we relied on archival sources rather than direct observation of board dynamics and the CEO dismissal processes. Second, the sample is limited to large U.S. firms, and thus our findings may not be generalizable to smaller U.S. firms or firms in other countries. In this regard, we suggest interesting avenues for future research. First, future extensions of our research may incorporate firsthand data (e.g., survey data or interviews with directors) that can more directly portray directors’ actual perceptions about different individual attributes and how potential faultlines are activated. Second, future research may replicate or extend our study by using samples of smaller firms or firms in other countries with different corporate governance systems from that of the large U.S. corporations used in this study. Such future research will advance our understanding of the impact of board faultlines on various outcomes, including CEO dismissal.

We argue that studying board faultlines can advance our understanding of the board decision-making process in general and the firm performance–CEO dismissal relationship in particular. Our findings suggest that demographic faultlines can disrupt important board decisions, while information-related faultlines can enhance the quality of decisions. Although boards are becoming more demographically diverse than ever before, they should expand their focus beyond highly visible demographic attributes and further diversify on nondemographic attributes. We hope our research will motivate future studies to examine the role of faultlines and group dynamics in boards and TMTs in other important firm outcomes.