APPENDIX
RESEARCH DESIGN, METHODS, AND SAMPLE
WHEN I BEGAN the research that became Searching for a Corporate Savior, I did not expect to learn what I have written about in the foregoing pages. As a doctoral student trying to describe the CEO labor market in terms of existing economic and sociological theory, I expected to find that the external CEO market was much like any other managerial labor market. For example, when I began my research on executive search firms, I believed that they would prove to be market brokers in the classical sense of the term—that is, parties that connected firms and potential candidates who were unknown to one another. I went into executive search firms seeking evidence to confirm this hypothesis. However, after several months of interviewing consultants and working with their research departments hoping to understand how a search firm uses its networks to find CEO candidates, while also extensively observing how directors and candidates behaved in the search process, I realized that the role of executive search firms was very different from what was commonly understood. It could not be adequately described with concepts such as matching supply with demand or filling structural holes. Something else was clearly at work in this and many other aspects of the external CEO search process.
What that “something else” proved to be is broadly illustrated by an interview with a corporate director that I conducted not long after completing my doctoral dissertation on CEO turnover in 1998, by which point I had already begun to suspect that the external CEO market was indeed (as I have put it in the title of chapter 2) “a different kind of market.” For many of the same social and cultural dynamics evident in the course of this single, relatively short encounter play themselves out on a larger stage when large corporations today go into the external market in search of a new CEO.
It was still dark when early one December morning I left my parents’ home in Ridgefield, Connecticut, where I had driven from Massachusetts the night before, to catch the 5:39 Metro-North train to New York City. The train arrived at Grand Central Station at 6:32, leaving me plenty of time before my 7:30 breakfast meeting. I walked from the station to the Waldorf Astoria Hotel, which—even though I grew up in New York City (albeit in Queens, which people from Manhattan will tell you is not really New York)—I had never set foot in before.
I was meeting with a director of three large, publicly held corporations who had agreed to be interviewed on the topic of CEO succession. I had been able to make an appointment with this individual through two contacts—one with a search consultant whose firm I had been researching, and another with a director whom I had recently interviewed. (The director who had referred me to the individual I was about to meet was someone I had met through a professor at the Harvard Business School.) Without these initial personal introductions, I would have never been able to secure my present appointment.
The director I was going to have breakfast with—whom I will refer to as John—was in New York on board business. Although I had never met John, I immediately recognized him when he walked into the Peacock Alley (the restaurant, now closed, that was then located just off the hotel lobby). His face was familiar to me from the annual reports and business magazine articles I had studied before our meeting. We sat down for breakfast. John, who had enough of a paunch to suggest that he probably enjoyed a good meal, nevertheless ordered a fruit salad with cottage cheese. I ordered the same, although I would have preferred eggs. In these kinds of situations, I had learned, it was best to follow the lead of the interviewee.
John’s persona exuded the kind of confidence one would expect from a person in his position. He looked directly at me when speaking. His voice was deep and assertive. He seemed like someone who had probably played football in college. And although it was December in New York City, he sported a healthy tan. We engaged in some social conversation. John was in town for a board meeting, but would be returning to his vacation home in Palm Springs later in the day. He already knew some things about me. Obviously, he had spoken to his director colleague, who had provided him with some background on my research. Most importantly, John told me that the referring director had described me as a “good guy.” In a social ritual with which I was becoming familiar, we then began trying to identify other people we both knew. John named a couple of the directors of the company I had worked for before graduate school in an attempt to identify a common acquaintance. I knew the names, but admitted that I had been at a level of the organization where I would not have been known to these individuals. We then spoke briefly about the director who had referred me to John. After a moment or two of discussion, we both agreed that he was a “really good guy.” Then, without missing a beat, John shifted the conversation to the matter at hand. “You are interested in CEO succession,” he said. “Let’s talk CEO succession.”
We spoke for nearly ninety minutes. During this time, John described to me the CEO successions he had been involved in, offering perspectives from both sides of the transaction. John himself had once been an outside candidate for the CEO position at a major corporation, a position he eventually was offered, accepted, and had recently retired from. One of the search firms I had been working with had placed him in this job. As a director of several large public corporations, he was now on the “demand side,” no longer a “supply-sider,” he joked. Having not only served on boards and search committees engaged in external CEO searches but also having headed a search committee, he had some significant understanding of the CEO firing and hiring processes. For the purposes of this interview, I tried to steer the conversation toward CEO hiring, especially outsider CEO hiring.
John’s descriptions of the searches he had been involved in were detailed. He began by pointing out that forced CEO successions were rare events but becoming more common. He discussed the pressures exerted on boards by Wall Street analysts and the business media, the “shortterm” perspectives of analysts and institutional investors, and how boards often responded to the demands of these constituents by firing the CEO. He then described the difficulty of finding “real leaders.” There was “a shortage of real leaders” inside most firms, John assured me. Moreover, getting one of the few “real leaders” in the corporate world to change jobs was both difficult and expensive. In discussing one of the searches he had been involved in, John described how directors “almost always already had some names in mind” for the position. The “big question” was whether these individuals would be interested in the position. In describing how boards learned which potential candidates were available and which were not, John spoke of the search consultants he had worked with in ways that were colorful, funny, and subtly scornful of the search profession in general. Finally, he described the process by which the board focused on the few final candidates, and why it eventually chose one candidate over the others.
Toward the end of our conversation, I asked John whether the process he was describing for selecting candidates identified the best person for the job. (I had not yet begun to form any opinion about this myself.) Obviously irritated by the question, he replied, “Of course it does.” Perhaps the fact that he had reached the CEO suite by means of this very process determined his response. At any rate, my question prompted him to reiterate the points he had made at the start of our discussion. CEO succession was a rare event, but the decision about who to appoint as CEO was the most important one that a board ever made. It was a decision that affected a number of constituencies, and their reactions to it had to be anticipated. John argued that the process was very dependent on individuals’ judgments, and therefore that much hinged on the quality of the judges. A board, he assured me, did not take this task lightly and was vigorous in pursuing any and all information about a candidate it could gather.
I took careful notes while John recapitulated many, if not most, of the points that I was beginning to see illustrated again and again the more I studied external CEO searches. Yet some of what I was learning had to do not just with what he and other directors I was meeting (along with candidates and search consultants relating their own sides of the story) were telling me but also with the social codes, behaviors, and rituals that I both observed and participated in during the course of our interview. The same kinds of rules that both enabled and shaped my encounter with John—rules about matters such as presentation of self, status and hierarchy, and what kinds of shared affiliations established a presumption of common values and permitted a measure of self-revelation—were, I was beginning to realize, absolutely essential to the functioning of the external CEO market. I didn’t know what kind of market it was or if it even was a market, but I was coming to believe that it deviated in profound and significant ways from existing theory and popular accounts.
Field Data Sources
Elites are notoriously difficult individuals to get access to. To get started with my field research, I relied on faculty members at the Harvard Business School to introduce me to search consultants and directors. Many Harvard Business School faculty are board members, and several of them offered to introduce me to fellow directors. I then used these initial connections to get access to other directors.
The forty directors I interviewed ranged in age from fifty-seven to seventy-two. Eighteen of them were either active or retired CEOs of Fortune 500 companies. Most were currently active members of the boards of large, publicly held companies. Many had also served on several other boards over the years. Several of these directors served on boards together. My estimates suggest that, via such director interlocks, I had indirect access to about one-fifth of the Fortune 1000 firms.
To prepare myself for the director interviews, I analyzed various documents about the firms with which these directors were associated. Among the most important were annual reports, 10Ks and 10Qs, and reports in the business press, especially from the Wall Street Journal, the New York Times, Fortune, and Business Week.
Typically, interviews with directors lasted from one to two hours. I tried to focus the interviewees on successions they had been involved in between 1990 and 2002. In total, forty separate succession events across fifty-five companies between 1990 and 2001 were discussed in depth.
When discussing CEO succession with directors, I found that the use of executive search firms came up in every discussion. I realized that I knew very little about search firms, and so soon after my first interviews with directors I began to look for an entry into this industry. Serendipitously, a Harvard Business School MBA classmate had done a summer internship at one of the large search firms and offered to introduce me to her executive sponsor there. At the same time, I was able to make contact with two other search firms. After several visits and phone calls, two of the largest search firms agreed to let me study their work in depth. Working with my search firm executive sponsors, I arranged meetings with the search consultants who focused on CEO searches for large corporations. While most of my meetings were in New York City, I also interviewed search consultants based in other locations including Atlanta, Chicago, and Palo Alto.
My interviews with search consultants typically lasted from ninety minutes to three hours. At the conclusion of the interview process, I had conducted thirty interviews with consultants and developed scores of mini-cases on CEO searches.1 In addition, I conducted follow-up phone and e-mail exchanges with several consultants to clarify issues as my knowledge of the industry and of the search process increased.
I experimented with tape recording and not tape recording interviews. Ultimately, my impression was that recorded interviews were more self-conscious and uncomfortable for the interviewee. I also found that by taking notes I was more involved and attentive to the interviewee. I took extensive notes during each of the interviews, which I typically transcribed within two days.
The interview guide I developed for studying both directors and executive search firms focused on three key areas: (1) the factors surrounding CEO succession, (2) the directors’ and search consultants’ exact roles during the CEO search, and (3) why a particular candidate eventually was selected. I encouraged the interviewees to talk about specific cases, not in generalities, when discussing the search process. In an attempt to capture as much detail about the search process as possible, I asked the interview subjects to focus on CEO searches since 1990 (about which their memories were likely to be better than about searches conducted before that date). I relied on an open-ended interviewing technique to discuss the CEO succession and selection process. With all my subjects, my approach was to carry on a partially guided conversation that elicited rich, contextual detail that could be used for analysis.
I supplemented my interview data with data from several other sources. I spent fifty hours conducting field observations of actual CEO and director searches. Some consultants allowed me to listen in on phone conversations with candidates and clients. Here I became informed about the perceptions of the various participants concerning the market and how well a particular search was proceeding. I was also permitted to directly observe and communicate with the search firm research staffs working on particular searches.
Field Data Analysis
For two years after completing my dissertation, I gathered additional field data on directors, candidates, search firms, external CEO successions, and the CEO labor market while also studying sources such as archival data and case files. I simultaneously began the process of reanalyzing my data with the goal of understanding the external CEO market using more general concepts. My approach was to separate what I had studied into two elements, content and form. I saw content as the interests and motives behind the actions of the participants in the market, and form as the specific process through which the interests and motives of the various participants were pursued. In reality, as several social theorists have noted, the two terms are inseparable and the distinction is only analytical (Simmel 1902; Weber, Roth, and Wittich 1978).
To get at the form of the phenomenon I mostly used techniques suggested by Strauss and Corbin (1990) to identify the phases common to all the searches I was studying. Based on the mini-case files I had compiled (see below), I wrote narratives summarizing each of the specific searches studied. I then broadly open-coded the various manifestations of participants’ understanding of the CEO labor market. I looked for shared sentiments, references, and perceptions in the interviews that extended beyond the specific cases being discussed. Then, when I found disagreement in, for example, search consultants’ descriptions of the process, I sought to understand the reasons for this deviation. Through this process of reconciling the general tendencies with the exceptional cases, theoretical concepts emerged through continued analysis as I sifted and resifted through my field notes, followed up with consultants, and acquired additional information about particular firms. It was during this time that I started to become aware of the form of the external CEO market with its defining features of small numbers of buyers and sellers, risk to participants, and concerns about legitimacy.
Next I began to identify the content of the interactions in the external CEO market. I analyzed the accounts given by interviewees regarding their activities during the search and the reasons given for these activities. In this analysis I took an iterative approach, going back and forth between the interview data and presenting my impressions and developing understanding of the process to colleagues. Further, during periodic updates about the progress of my research, I discussed my findings and thoughts with my executive sponsors from the search firms and with directors.
Statistical Data
DATA AND METHODS
Sample. The statistical data used in this book consists of the Fortune 500 for 1980 plus the 100 largest commercial banks, 100 largest financial services firms, 100 largest retail firms, and 50 largest transportation firms. I follow each of these firms through 1996. Executive compensation data was collected from 1991 to 2000. While the selection of large corporations limits the generalizability of my results, I decided on these firms because they are widely followed in the business media, which in turn offers more complete information on company events than is available for smaller firms.
Exhaustive data on the CEO changes within these firms was collected for the period between 1980 and 1996. These successions include all successions between 1980 and 1996 that satisfy the following criteria: (1) the incumbent and successor both can be identified, (2) the Wall Street Journal, New York Times, or Business Week reported the succession announcement, (3) characteristics of the CEO’s tenure with the firm and in the position could be collected, and (4) the succession was not directly related to a takeover. These criteria were met in all cases.
Forbes’ annual Executive Survey, Dun and Bradstreet’s Reference Book of Corporate Management, Standard and Poor’s Register of Corporations, and firm proxy statements (10Q and 10K) are used to collect information on CEO tenure, years with the firm, and age. The exact date of and reason for each succession is obtained from the media announcements.
Financial data for the firms was collected from Standard and Poor’s COMPUSTAT database.
Board of director data was compiled from Dun and Bradstreet’s Reference Book of Corporate Management, Standard and Poor’s Register of Corporations, and firm proxy statements (10Qs and 10Ks).
Executive compensation data was collected from Standard and Poor’s Execucomp database.
DEPENDENT VARIABLES AND METHODS
CEO Firing. For the analysis of CEO dismissals presented in figure 3.3, I studied the event of forced dismissal. I created two discrete categories of turnover: natural turnover (coded 1) and forced turnover (coded 2). The baseline was no turnover. Therefore I used a multiple-destination event-history model. Forced turnover includes cases in which a CEO departs before age sixty-one and does not leave for an equivalent position at another firm. Natural turnover consists of cases in which a CEO departs for another job or leaves because of retirement, illness, or death.
The logic of the model assumes three possible outcomes for CEO changes: (1) CEOs who do not change their jobs; (2) CEOs who voluntarily move to another job or reach the retirement age; and (3) CEOs who involuntarily leave their positions. The different destination states for CEO transitions require us to model these CEO events as competing risks:
Tvt = (Tvt*)[fvt(rvt)]
Tit = (Tit*)[fit(rit)]
where v and i refer to voluntary and involuntary turnover, and where each transition (T) is modeled conditional on the competing event not having occurred (Blossfeld and Rohwer 1995). Each of these competing events involves a discrete change of state. These changes can occur at any point in time and the model suggests that there are both time-constant and time-varying factors influencing these events.
The type of analysis used to model the competing events given both changing and constant independent variables is continuous-time transition rate analysis with time-varying co-variates. Effects on the transition rate are estimated by maximum-likelihood procedures that model the probability that a CEO moves out of a job given a set of co-variates. In other words, transition rates are conditional probabilities for the occurrence of the discrete events of interest (the rates of voluntary and involuntary turnover). My main interest lies in how dependent these events are on the set of independent variables I have identified.
As Tuma and Hannan (1984) point out, a number of different continuous-time stochastic models can be specified for the transition equations specified above. I used the log-logistic distribution to model the time dependence of the CEO transition process since this model explicitly models CEO changes in terms of waiting times. To examine the appropriateness of this model, I compared the maximized log-likelihoods assuming this model against those for the exponential, Weibull, and log-normal models; I found that the log-logistic model provided a better fit to the data (significant <.01). A visual and regression test of the parametric assumptions suggests that the log-logistic model provides the best linear transformation. A regression test of the parametric assumptions suggests no substantive difference between either the log-normal model or the log-likelihood. The fully specified model with its logarithmic time-dependence is specified as:
where r is the transition rate from origin state j to destination states k. The associated co-variates, a(jk) and b(jk) are the model parameters to be estimated.
Outsider Succession. An outside CEO appointment is defined as one in which the new CEO assumes the CEO title within one year of the date that he or she joins the firm. I classify CEOs who join the firm as long as one year before their appointment as outsiders because new CEOs who have been employed at the firm for only one year are likely to have been hired with the expectation that they would eventually be appointed to the CEO position. Hiring an executive before the CEO succession allows him or her to obtain additional specific human capital necessary for the position. The above definition of outside CEOs is consistent with the wide range of definitions used in other studies. For instance, Reinganum (1985) classifies executives who join the firm at the time of the succession as outsiders, while Vancil (1987) includes all executives who have been employed at the firm for five years or less. In my sample, 27 percent of the successions involve outsiders.
To study the probability of outsider succession, I use both logit and multinomial logit estimations for analyses of the turnovers. I use logit modeling to understand the factors that affect the selection of insider versus outsider CEO candidates. I use multinomial logit modeling because I have a dependent variable that can take on more than two discrete outcomes and these outcomes have no natural ordering (i.e., no turnover, insider selection, outsider selection).2 Both models are estimated as:
where j is the number of outcomes and x is the estimator(s). If there are only two outcomes (j = 2), estimates are identical to those produced by logit modeling (Greene 1993: chap. 21).
INDEPENDENT VARIABLES
Tenure. CEO tenure is dated from the year the individual at the firm under consideration took the CEO position. Tenure was determined as the difference between the last year a CEO occupied the position and the start year, plus one. If the CEO continued to occupy the position in 1996, the observation was treated as censored. To avoid left-hand censoring, I coded the original start years of all the CEOs as of 1980. For example, Lee Iacocca, CEO of Chrysler in 1980, was originally appointed in 1978. Consequently, his start date is coded as 1978. As a result, I was able to measure the tenure of each CEO in the sample and, therefore, avoid the potential issue of survivor bias for subsequent CEO appointments. Tenure serves as the time clock for the event-history tables—that is, a CEO’s tenure defines a “spell.”
Firm Performance. I use annual operating returns of the firm to proxy the CEO’s performance. The annual operating return for a firm is defined as the ratio of operating income before depreciation and taxes to operating assets. Because operating income does not include taxes, royalties, dividends, interest income received, or any dividends paid to stockholders, it is considered a robust measure of the operating performance of an organization (Smith 1990).
Founder. I coded a dichotomous variable to indicate firms in which the CEO is also the founder of the company. Founder CEOs are coded as 1, non-founder CEOs as 0. This variable proxies for the power that founder CEOs derive from their close identification with the company.
Separation of Chairman/CEO Positions. I coded a dichotomous variable to indicate firms in which the CEO and chairman positions are separate. The separation of the positions proxies for the relative control the CEO can exercise over the board’s access to information and resources. Firms in which the positions are separate are coded as 1. Firms in which the two positions are held by the same person are coded as 0.
Year of Hire. Dummy variables for the period during which the CEO was hired are included in the model. These dummy variables capture both the effects of the social conditions during the period within which the CEO was hired and the effects of tenure. For the firing analysis, three categories were used: CEOs appointed before 1980, and CEOs appointed after 1985. The omitted category consists of those CEOs appointed between those two periods.
Director Composition and Interlocks. I compiled data on board of director composition and board interlocks. Inside board members were those directors identified as current or former executives of the firm. Insiders were coded as 1, outsiders as 0. I created two measures from this data. First, I measured the total number of directors. I also created a percent-insiders measure to capture the relative strength of outside directors.
The “centrality” of a firm’s board members to other firms is used to measure the embeddedness of the board within the interlocking directorate. When defining a network it is best to be as inclusive as possible in determining which elements of the network to include, particularly when the diffusion of information and practices among contacts is at issue; thus, I compiled board data for all firms required to report board composition to the SEC. This includes all firms in the sample. Inside board members are those directors identified as executives, former executives, or others financially connected (e.g., affiliated law firm or consultant) to the firm.
Board interlocks are defined as the sum of all ties a firm’s board had to all other boards in this sample, that is, the total number of other boards each director sat on, summed across all directors (minus any redundant ties). As discussed in chapter 4, both directors and search consultants believe that the board of directors is a primary source of specific information on potential candidates. In a direct attempt to more closely operationalize the concept of how some boards would have better information than others, I asked interviewees, “How could this be known?” Both the search consultants and directors consistently responded that the ability of the board to access other directors who could provide more detailed information about the candidate mattered. In particular, the directors and other CEOs they knew from board connections were important.
Because the interviewees suggest that the type of information necessary to make a CEO hiring decision is confidential and emerges from director connections, I include two measures of centrality. The first measure used is degree centrality, which is simply the number of interlocks minus any redundant ties. It considers the direct connections a firm has to other firms (Davis 1991; Haunschild 1993).
The absolute number of firm interlocks has been used to operationalize several concepts in the organizational literature. As discussed earlier, Davis (1991) and Haunschild (1993) have used number of interlocks as a measure of normative and mimetic influence, respectively, on a firm’s board. Others have used it as a general information measure (Useem 1984). This measure has also been used by organizational economists to argue that heavily interlocked boards are often too busy to govern effectively. Core, Holthausen, and Larcker (1997, 16–17), for example, argue that large numbers of interlocks reduce a board’s ability to attend to its duties because of time constraints and thereby reduce board effectiveness.
A number of criticisms, however, have been directed at relying on this measure alone in director interlock studies. First, it is a proxy for several network concepts and, therefore, limited in its explanatory power for any single concept (Mizruchi and Bunting 1981). Second, the measure ignores the variance in the number of directors per company and, consequently, the potential of a firm to establish connections (Bonacich 1987). Finally, it weighs all interlocks equally and assumes a linear relationship between interlocks and other organizational characteristics, such as size (Mizruchi and Bunting 1981).
Thus, to isolate the information effects of director interlocks suggested by the field research, I use a second measure that has become more prominent in the network research to capture the prominence and depth of involvement of actors in relations with other actors—the Bonacich centrality measure. This captures the centrality of an actor by considering not only direct or adjacent ties of a firm’s directorate but also indirect paths involving intermediaries. Thus, the Bonacich measure would capture the difference between a firm’s being connected to IBM (a computer firm with many ties to other firms) versus connected to Grand Union (a supermarket chain with only one tie to other firms). The Bonacich measure includes this possibility by incorporating a parameter that reflects the extent to which an actor’s centrality is a function of the centrality of actors to whom the actor is tied. This measure can be considered as an information indicator (Bonacich 1987; Hansen 1996). That is, better information is likely to be obtained from directors who are in contact with prominent or central directors who themselves are in contact with many other directors. The strength of this measure for the study of director interlocks has been described in several methodological reviews and empirical studies (Mintz and Schwartz 1981; Mizruchi 1982; Mizruchi and Bunting 1981). Formally,
where N = number of organizations linked with i, rij equals the number of connections with the particular link, Cj equals the centrality of the organizations linked with i, and j is the largest eigenvector of the matrix r.
Because the two measures of centrality and Bonacich power are highly correlated and potentially likely to bias the standard errors, I use the residuals of Bonacich power as the measure.3 This involves formalizing the relationship to model the overlap between the two centrality measures and using the residuals as the predictor. This measure separates out the degree-centrality effects from the information-centrality effects. This is a standard technique when correcting for multi-colinearity (Kennedy 1992). The assumption here is that multi-colinearity arises from an actual approximate linear relationship among some of the regressors, which in the case of the Bonacich measure and number of interlocks is mathematically true. Consequently, this relationship is formalized and the estimation then proceeds in the context of a simultaneous equation-estimation problem.