Upload
lykiet
View
226
Download
2
Embed Size (px)
Citation preview
Internal vs. External Promotions: Evidence from Professional Football Coaching Careers*
C. Edward Fee Michigan State University
Charles J. Hadlock Michigan State University
Joshua R. Pierce Michigan State University
This Draft: November 8, 2002 * We thank the staff at the Pro Football Hall of Fame in Canton, Ohio for many helpful conversations and for allowing us access to their historical archives. The most recent version of this paper can found at http://nebula.bus.msu.edu:8080/feehadlock. All errors remain our own.
Abstract
We examine the job movements of professional football coaches both within and across employers and compare the mechanisms governing internal and external promotions. We find that the likelihood of an external promotion is strongly related to individual performance measures and only weakly related to team performance. In contrast to external promotions, we find that the overall likelihood of an internal promotion is unrelated to an individual’s performance. This difference between internal and external labor markets appears to arise from the process governing job openings within the internal hierarchy, as the likelihood of an internal job opening up is negatively related to performance. Conditional on an internal opening occurring, we do find that increases in individual performance increase the probability of being promoted. Relationships matter a great deal in this labor market in the sense that coaches appear to be fired and hired as a group, suggesting that the value of an individual to an employer depends on the identity of the entire set of individuals who work together. We argue that our findings have implications for several issues related to incentives and organizational design.
JEL Classification: D23; J41; L22; L83 Key words: Promotions; Internal Labor Markets; Hierarchies; Performance Evaluation; Careers; Football; Teams
1
1. Introduction
Economists have long recognized that there may be important differences between
transactions that occur within a firm and transactions that take place in external markets. One
important area where these differences may manifest themselves is in the allocation of
employees to jobs. Consequently, recent empirical work has devoted considerable attention
towards understanding how internal labor markets function. These studies have identified
several interesting empirical patterns that characterize career profiles within an employing
organization. However, because of data limitations, many of these studies use data from a single
employer and ignore the role of the external labor market in the evolution of careers.
In this paper we directly compare and contrast internal and external labor market
transactions by constructing a dataset on the job movements of professional (American) football
coaches both within and across employers. In our analysis we focus primarily on promotions
from level-two positions (i.e., offensive and defensive coordinator positions) to level-one
positions (i.e., head coach positions). This labor market is ideal for our purposes since career
movements can be identified from public sources. In addition, we are able to construct detailed
objective performance metrics that measure the performance of both the organization as a whole
(team performance) and the specific activity overseen by the individual (offensive performance
or defensive performance). Since the positions we examine are very senior positions and the
industry is characterized by a great deal of inter-firm movement, our results are most likely to
generalize to other labor markets with these key characteristics. We suspect that one such
market is the market for senior managerial talent at publicly traded firms.
2
Our first empirical goal is to determine whether internal and external labor markets differ
in the mechanism that they use in choosing individuals for promotions. After identifying some
important differences, we then attempt to determine the characteristics of the internal labor
market that result in the observed differences. While much of our focus is on the differences
between internal and external labor markets, our findings also add to the existing body of
knowledge concerning how each of these labor markets function. In particular, in contrast to
almost all prior studies, we are able to look directly at the role of objective measures of an
individual’s performance in the internal and external assessment of that individual’s ability. We
are also able to examine the tournament nature of internal promotions, since typically a firm has
two individuals occupying a level-two position vying for a single level-one spot.
When we estimate models predicting external labor market promotions, we find that
individual performance measures are significantly positively related to the likelihood of a level-
two individual being hired to a level-one position at a new employer. In contrast, individual
performance measures have a negative and insignificant effect on the likelihood of an internal
promotion. This evidence indicates that the incentives generated by promotion prospects near
the top of a hierarchy may be driven primarily by external rather than internal career
opportunities.
To examine why internal promotions are not related to individual performance metrics,
we consider the dynamics of the promotion process. For a given individual to be promoted
internally it must be the case that (a) there is an opening for an insider, and (b) the firm chooses
the individual over other internal candidates. It turns out that the process governing openings is
negatively related to an individual’s performance, while the process governing being chosen for
the job conditional on an opening is positively related to an individual’s performance. Since the
3
two effects roughly cancel out, there is no positive relationship between individual performance
and internal promotions. This evidence suggests that the notion of slot constraints and job
congestion near the top of an organizational hierarchy needs to be taken seriously in thinking
about careers.
With regards to external labor market promotions, we present evidence that the labor
market’s inferences concerning an individual’s ability depend on individual rather than team
measures of performance. In particular, while individual performance metrics are significantly
related to the likelihood of an external promotion, team performance appears to have no
independent effect on this likelihood. This finding may have interesting implications for whether
organizations should report disaggregated performance data (e.g., division-level performance
metrics in a multi-division firm) that may allow the labor market to more effectively poach their
best employees.
While most of our focus is related to the promotion process, we also have some data on
lateral movements of coaches across teams. Our evidence here indicates that most lateral
movements reflect a process where level-two coaches are let go following poor team
performance. Within this set of laid off individuals, those with the best individual performance
tend to be able to secure comparable positions at a new employer. It also appears that there is a
significant amount of relationship-specific human capital in this labor market. When head
coaches are fired, their level-two subordinates very frequently also find themselves out of work.
When head coaches are hired, they tend to bring along individuals that they worked with in the
past.
The rest of the paper is organized as follows. In section 2 we discuss the relevant
literature and motivate our empirical investigation. In section 3 we discuss our data and sample
4
selection procedures. Our main evidence on differences between internal and external
promotions is presented in section 4. We investigate the dynamics of internal job openings and
promotions in Section 5. In section 6 we examine lateral job movements and some evidence on
relationship specific human capital. Section 7 concludes.
2. Upward mobility in the labor market
2.1 Theories of internal labor market promotions
One set of theories regarding internal promotions focuses on the role of promotions as
pure incentive devices (e.g., Lazear and Rosen (1981)). The insights from this literature suggest
that firms may use promotions as a way of committing to an incentive scheme that elicits
appropriate effort from their employees.1 While promotions may very well generate incentives
for agents to take actions that help them move up the hierarchy, several authors have pointed out
that promotions may not be the best way to solve a pure incentive problem (e.g., Baker, Jensen,
and Murphy (1988)). In particular, since promotions involve a change in an individual’s job
responsibilities, it is not clear that using promotions for pure incentive reasons will generate the
optimal assignment of individuals to jobs.
A second set of theories regarding internal promotions abstracts from incentive
considerations and focuses on the role of promotions as an optimal assignment mechanism.2
Underlying many of these models is the basic idea exploited by Rosen (1982) that more talented
individuals should be assigned to higher positions in a hierarchy. Talent can have many
1 For more on tournaments, see Rosen (1986), Main, O’Reilly, and Wade (1993), and Bognanno (2001). 2 For an overview of this literature, see Gibbons and Waldman (1999b) and the references cited therein.
5
dimensions, including an innate ability component that is learned by employers over time, and a
human capital component that grows with experience. Optimal assignment considerations would
then suggest that individuals are promoted when information concerning their ability and/or
skills developed on the job dictate that they are optimally assigned to a more senior position.
Gibbons and Waldman (1999a) present a model of promotions and wage dynamics inside of
firms that incorporates these elements and is consistent with some of the existing evidence on the
promotion and wage process observed in firms.
It is worth noting here that while internal promotions in firms may be structured primarily
to optimally match individuals to jobs, if the job assignment process is affected by employers’
beliefs concerning innate ability, there is still likely to be an incentive for individuals to take
actions that affect the ability inference process (see Fama (1980) and Holmström (1982)). Thus,
even if promotions are not specifically designed for incentive purposes, the mechanism
governing promotion decisions could have important incentive effects. One of the goals of our
study is to understand exactly what these incentive effects are.
An issue that is related to the role of promotions as an optimal assignment mechanism
arises from the possibility that in some employment settings, particularly near the top of a
hierarchy, the number of jobs of a particular type or level may be fixed by technological or
organizational considerations. For example, a firm typically can have only one Chief Executive
Officer. In these settings where there are “slot constraints” on jobs, the internal promotion
process will be linked to the mechanism governing job openings.3 An individual’s superior must
depart in order for the individual to have a chance of getting a higher-level position. If job
openings rarely arise when an organization is performing well, there may be very little chance of
3 Most models of job assignments with slot constraints are static. See Rosen (1982), Waldman (1984a), and MacDonald and Markusen (1985).
6
a subordinate being promoted following strong performance.4 Thus, the incentive effect from
internal promotions may be weaker than one would expect in a setting where job openings arise
exogenously or where jobs can be created to appropriately match an individual’s talents to a
position.
2.2 Empirical evidence on internal promotions
Several studies have examined the internal promotion process.5 Since wage changes are
often large when an individual is promoted (see Murphy (1985) and Baker, Gibbs, and
Holmström (1994b)), it is clear that incentives generated by promotions can be substantial. In
examining the internal promotion mechanism, several findings in previous studies are consistent
with the hypothesis that individuals exhibit substantial heterogeneity in innate ability and that
inferences concerning ability affect job assignments. In particular, existing evidence suggests
that individuals who were most recently promoted are more likely to be promoted again,
suggesting that a long tenure on the current job is a proxy for low ability (i.e., a long tenure
indicates that an individual has been passed over). There is also evidence that other measures of
ability, including past wage growth and subjective performance measures, affect the likelihood
of promotion.
While the existing literature on the promotion mechanism establishes many interesting
findings, there are still some important missing links. First, for reasons of data availability,
many of the existing studies are limited to a single firm. Consequently, the generality of their
4 A related issue is whether the firm decides to choose an insider or an outsider when there is a job opening. If firms tend to choose outsiders, this will have a deleterious effect on internal promotion incentives. For an analysis of this type, see Chan (1996). 5 See Medoff and Abraham (1980), Lazear (1992), Blackwell, Brickley, and Weisbach (1993), Baker, Gibbs, and Holmström (1994a), and Gibbs (1995).
7
findings is unclear. Second, the relationship between objective measures of individual
performance and internal promotions has not been firmly established. This is important, since
there are alternative interpretations to some of the existing findings. For example, individuals
who receive high subjective performance marks from superiors may not be promoted because
they are the most able. Instead, firms may choose who they want to promote for other reasons
and then assign them high performance marks. If we can establish that a high level of objective
performance leads to an increased likelihood of an internal promotion, we will have stronger
evidence that promotion policies are influenced by learning and optimal job matching
considerations. In addition, this evidence would indicate that individuals have an incentive to
increase their observed performance to increase their likelihood of promotion.
A third issue that is not addressed in existing empirical studies is the role of slot
constraints in the promotion process and in promotion-based incentives. This could be a
particularly important issue near the top of a hierarchy, as pointed out by Baker, Jensen, and
Murphy (1988). For high-level employees, the nature of promotion-based incentives is difficult
to assess without understanding the process governing (a) job openings, (b) the decision to hire
an insider, and (c) the decision of which insider to hire. We hope to shed light on these issues in
our analysis.
2.3 The external labor market
Many of the factors that govern promotions within the internal labor market should also,
at least in theory, govern promotions in the external labor market. In particular, if the ability of
an individual is revealed to be high, it may be optimal for the individual to switch to an employer
8
who can more efficiently use the individual’s talents. Consistent with this hypothesis, Fee and
Hadlock (2002a) present evidence that high-level executives at superior performing firms tend to
jump to better jobs at new employers.
While similar economic considerations may govern internal and external labor market
promotions, there are several reasons to expect there to be differences in these promotion
mechanisms. First, for positions with a substantial amount of firm-specific human capital, it is
unlikely that individuals will find their services to be worth more outside the firm than they are
worth to the current employer. Thus, the overall rate of internal promotions may substantially
exceed the rate of external promotions. Second, for many positions it is likely that the current
employer has different information regarding the employee than does the outside market.6 Thus,
the performance metrics that predict internal promotions may vary from those that predict
external promotions.
A third potential difference between internal and external promotion mechanisms arises
from the observation that promotion opportunities in the external labor market are likely to arise
for reasons that are independent of how an individual is performing, while internal job openings
may not exhibit this independence. For example, the number two individual at a unit that is
performing well may be very unlikely to be promoted internally, because the chances of his
superior leaving are particularly low during times of strong unit performance.
A final potential difference between internal and external promotion mechanisms arises
from the fact that internal labor market transactions occur in the context of a firm’s implicit and
explicit contractual relationship with all of its employees, while external promotions do not. For
example, reputational or incentive design considerations may lead a firm to promote an internal
6 In fact, this difference in information motivates several theoretical studies of job assignment. See, for example, Waldman (1984b) and Lazear (1986).
9
candidate who was previously promised the promotion, even though the candidate is not the best
person to fill the job. In contrast, if an outside firm were looking to hire away an individual from
another firm, presumably they would simply choose the most talented individual.
2.4 Empirical strategy
Given the arm’s length nature of external labor market transactions, our empirical
strategy is to first examine the mechanism governing external labor market promotions in the
market for professional football coaches. This analysis allows us to examine the role of
objective performance metrics and other variables in outsiders’ assessments of an individual’s
abilities. This analysis is of independent interest in the sense that it informs us on the role of the
outside labor market on incentives and allows us to test the theory of Fama (1980) and
Holmström (1982) on career concerns incentives induced by external labor market
considerations.7 In addition, this analysis allows us to establish a baseline of what labor market
promotions look like in the absence of a long-term employment relationship.
After establishing this baseline, our strategy is to estimate models predicting internal
promotions. In particular, we are interested in the role of objective measures of performance and
other important individual characteristics (e.g., age and tenure) on the likelihood of an internal
promotion. Viewed in isolation, this analysis allows us to assess the role of internal labor market
rewards on incentives. Viewed in the context of our findings on external promotions, this
evidence allows us to identify any substantial differences between internal and external labor
7 Note that Fee and Hadlock (2002a) have performed a similar analysis in the market for senior executive talent. In contrast to that study, in this study we have both aggregate and individual measures of performance. Thus we are able to more closely contrast and examine the metrics the external labor market uses in its assessment of an individual’s ability.
10
markets in promoting key personnel. For the differences we identify, we are interested in the
extent to which these differences can be attributed to the considerations outlined in the previous
subsection. Our data allow us to look most closely at the possibility that internal slot constraints
and the process governing job openings leads to differences in internal and external promotion
mechanisms and incentives.
3. Data and sample selection
3.1 Sample selection
We assemble a database of the names of all individuals employed by teams in the
National Football League (NFL) from 1970-2001 with the title of head coach, offensive
coordinator, or defensive coordinator. We choose to examine a sports labor market because of
the publicly available data on the identities and performance of key personnel. 8 Our specific
choice of the NFL coaching market is motivated by the observation that this labor market tends
to have fairly fixed promotion paths where offensive and defensive coordinators are often
promoted to head coach positions either internally or by switching employers. We begin the
sample collection in 1970, as this is the start of the first decade that follows the merger of the
NFL and the American Football League (AFL).
We record the identities and backgrounds of the individuals serving in the three identified
coaching positions as of the start of each season. Most of this information is identified from
reading annual media guides published by NFL teams and stored in the archives of the
8 See Rosen and Sanderson (2000) for a discussion of the benefits of using sports labor markets as laboratories to evaluate economic hypotheses.
11
Professional Football Hall of Fame in Canton, Ohio. We supplemented this information when
necessary with press clippings in the Hall of Fame library and with news searches on the Dow
Jones Interactive electronic database.
Before describing some of the sample characteristics, we describe here the organization
of the coaching function of a typical NFL team. All teams have a fairly hierarchical structure,
with the team’s head coach being the individual who is ultimately responsible for all decisions
made on the playing field. We will occasionally refer to the head coach position as a level-one
position. Head coaches are hired by a team’s senior management, typically by the team’s
General Manager. Senior management generally reserves the right to make the final decisions
on all player and personnel choices. However, the head coach generally has a great deal of input
into decisions regarding the employment of other coaches who report to him, and some input into
decisions regarding which players the team will employ.9
All teams have a set of more junior coaches who are subordinate to the head coach.
Since (American) football is a game where offensive play and defensive play are quite distinct
and entail using a different set of players, teams generally organize their junior coaches into a set
of offensive coaches and a set of defensive coaches.10 At the head of the offensive (defensive)
function is either the head coach himself or an offensive (defensive) coordinator. By the end of
the sample period, 30 of the 31 NFL teams had an offensive coordinator and all 31 teams had a
defensive coordinator. In the earlier sample years the head coach often acted as his own
9 See Wolf and Attner (1999), ch. 2, for an NFL General Manager’s perspective on hiring decisions. 10 On the offensive side, in addition to the offensive coordinator, teams frequently employ offensive line coaches, quarterback coaches, and receiver coaches. On the defensive side, in addition to the defensive coordinator, teams often employ defensive line coaches, linebacker coaches, and secondary coaches. Many teams also have coaches that cannot be categorized as purely offensive or defensive, for example special teams coaches and scouting coaches.
12
offensive and/or defensive coordinator.11 We will refer to the offensive and defensive
coordinators as level-two coaches, while all other more junior coaches will be referred to as
level-three coaches.
We report some basic summary sample statistics in Table 1. Our sample contains 901
franchise-years representing 2,284 person-years and 369 unique individuals. We identify a head
coach for all 901 franchise-years, while an offensive coordinator is present in 607 (or 67.4%) of
these 901 franchise-years, and a defensive coordinator is present in 776 (or 86.1%) of the 901
franchise-years. As the figures reveal, there has been a slight upward trend in number of teams
over the sample period (from 26 to 31), and a large upward trend in the propensity of teams to
employ offensive and defensive coordinators. The figures in the table indicate that the coaches
we identify are typically in their upper 40s and have worked for over 10 years in professional
coaching, suggesting that for these individuals coaching is a long-term career choice.
3.2 Typical Career Paths
As we discuss in section 2, our primary empirical goal is to estimate models of how
level-two coaches obtain promotions to level-one positions by moving within or across
employers. Before conducting an analysis of these job movements (i.e., an analysis of what
happens to our sample coaches), some relevant information can be gleaned from the background
biographical data on each coach (i.e., information on how they got to where they are). To
11 This change in organizational structure appears to be driven by the increased complexity of professional football strategies, for example the advent of the “West Coast Offense” and the “Zone Blitz”. For more on these technological considerations, see Lamb (1999) as well as other articles in Total Football II: The Official Encyclopedia of the NFL (1999).
13
organize this information, in Table 2 we treat each person-position match as a single observation
and report data on the previous position the individual held before the current assignment.
Interestingly, as the data in Panel A of Table 2 reveal, the majority of our sample coaches
worked for a different NFL team immediately before getting their current position. For all three
types of coaches, the fraction of individuals who were internally promoted to the current position
is less than one third. The rest of the individuals were hired externally, in the majority of cases
from other NFL teams, but in a small minority of cases from other professional leagues or the
college ranks. These non-NFL sources of coaching talent appear more important when firms
hire head coaches than when firms hire offensive or defensive coordinators. The data indicate
that approximately 1 in 4 outside head coach hires come from non-NFL sources, while less than
10% of level-two coaches are hired from outside the NFL.12
In Panel B of Table 2 we look at the previous positions held by individuals who obtained
their current position without switching employers. We find here that for head coaches who
were promoted internally, approximately half were promoted from level-two coaching positions
and approximately half were promoted from level-three coaching positions. While this may
suggest that some individuals “leapfrog” over their superiors from level three to level one, this is
in fact not the case. Almost all of these level-three-to-level-one internal moves are in early years
when many teams did not have a complete set of level-two coaches.13 In these cases, a level-
three-to-level-one promotion should be roughly equivalent to a level-two-to-level-one
12 When head coaches are hired from outside the NFL, they almost always come from head coach positions in other professional leagues or the college ranks. The only exception in our sample is Hank Stram who joined the Dallas Texans (later renamed the Kansas City Chiefs) as head coach in 1960 after an assistant coaching stint at the University of Miami. 13 Note that since in Table 2 we are using biographical data for each position-person match, we are including data on the prior employer of head coaches in 1970 who were promoted many years earlier, say in 1960. The earlier the date of the hire the more likely the team did not have a full set of level-two coaches. For the set of internal head coach replacements after 1970, only 5 represented “leapfrog” situations in which a level-3 coach became a head coach for a team with a previously designated coordinator on his side of the ball (offensive or defensive).
14
promotion. When we examine level-two coaches who were assigned to their current position
without a change in employer, the figures in Panel B of Table 2 indicate that almost all of these
individuals were promoted from a level-three coaching position with the same emphasis
(offensive or defensive) as the position they are promoted into.14
In Panel C of Table 2 we report the previous positions held by individuals who were
hired into their current position from another team. As can be seen in the table, slightly more
than half (55.26%) of externally hired head coaches were head coaches at a different employer,
while the majority of the others were level-two coaches elsewhere.15 In the case of level-two
coaches who were hired externally, the figures in Panel C of Table 2 indicate that approximately
80% were previously either level-two or level-three coaches in positions with the same emphasis
(offensive or defensive) as the position they were hired into. Note that some level-two coaches
were previously employed elsewhere as head coaches, suggesting that downward moves or
demotions in the external labor market are occasionally observed.
Taken as a whole, the figures in Table 2 paint a picture of a labor market with substantial
inter-franchise mobility and a great deal of outside hiring. In the internal labor market there are
some obvious promotion tracks, while demotions appear rare. In the external labor market, there
is ample evidence of job changes that appear to be promotions, lateral moves, and demotions.
With regards to individuals with offensive versus defensive experience, there is no obvious
preference to hire (internally or externally) head coaches with one type of experience over the
other. However, it is clear that level-two positions are filled primarily with individuals with
14 The one case that may appear from Table 2 to be an internal demotion into a level two position is where the head coach of the San Diego Chargers in 1970 (Charlie Waller) became the offensive coordinator in 1971. However, his status as head coach in 1970 was only a temporary assignment that was made because the previous head coach became ill. As in Baker, Gibbs, and Holmström (1994a), outright demotions in the internal labor market appear very infrequent. 15 We suspect that externally hired head coaches who were previously level-three coaches elsewhere were generally very senior level-three coaches.
15
prior experience that emphasizes the same activity (offensive or defensive) as the position that is
being filled. Consistent with some of the modeling by Gibbons and Waldman (2002), these
patterns are consistent with the notion that individuals develop job-specific human capital with
experience, and this job-specific capital affects what types of jobs individuals are promoted into.
3.3 Categorizing job changes
We now turn our attention to examining the circumstances surrounding job changes in
our panel. Since our emphasis is on the promotion of level-two coaches, we first discuss how we
categorized the labor market movements of these individuals. For each observation where a
person was in a level-two coaching position at the start of a season, we evaluate whether the
individual was still in the same position at the start of the subsequent season. If the individual
was in exact same position with the same team, we refer to this as a “no change” observation. If
the individual was promoted to a higher-level position with the same team, we refer to this as an
“internal promotion” observation. All cases in our sample where level-two coaches stay with
their team but experience a change in job title are internal promotions.
In instances where level-two coaches separate from their team, using our panel of data we
are able to identify whether the individual shows up at another team in a level-one or level-two
coaching position. When the individual shows up as a head coach at a new team within one year
of the start of the season where we identify the separation, we refer to this as an “external
promotion.”16 When the individual shows up as a level-two coach at a new team within the same
16 We use this timing convention because we are concerned that individuals may voluntarily separate from their old team at the end of a season and then take a short time to evaluate their labor market opportunities. As long as the individual shows up at a higher-level position within approximately one year of his job separation, we categorize the change as an external promotion. Given the thinness of this labor market and the fact that most hiring and firing is
16
window of time, we refer to this as a “lateral” (no pun intended) move. All other cases are
referred to as “dismissals.”17
The summary statistics for each of these five types of changes for our set of level-two
coaches is reported in Panel A of Table 3. As is evident from this table, the probability that a
level-two coach stays in the same position for an entire year is in the 65-70% range, implying a
great deal of turnover. The likelihood that a level-two coach is promoted internally is 2.57% for
offensive coordinators and 2.68% for defensive coordinators. When level-two coaches depart,
the most frequent outcome is that they do not show up at a new employer (i.e., an outcome that
we would call a dismissal). However, in a substantial minority of cases these individuals do
move to comparable or superior employment at a new team. The overall likelihood that a level-
two coach moves laterally to a new employer in any given year is approximately 9% for both
types of coordinators. The corresponding likelihood of being externally promoted to a head
coaching job elsewhere is 3.40% for offensive coordinators and 2.68% for defensive
coordinators. Based on these figures, the likelihood of an external promotion in any given year
for a level-two coach appears roughly equal to the probability of an internal promotion.
Since part of the process governing internal promotions of level-two coaches will depend
on the process governing the departure of level-one coaches, we also categorize the job changes
of head coaches in our sample. Similar to our earlier labeling, if a head coach is serving at the
start of one season and is still there at the start of the subsequent season, we refer to this as a “no
change” observation. We refer to all observations where the head coach changes as
done over a short window of time between seasons, this timing convention seems to be a reasonable way to identify cases where the external labor market rewards an individual because of a positive assessment of his talents. 17 Most coaches that show up at a new team do so fairly quickly. We have experimented with other timing conventions in defining these variables with regards to the window over which we look for new employment, and the results are very similar to what we report in the text. Note that if a level-two coach shows up at a level-three position elsewhere, we cannot track these movements and will end up categorizing this as a dismissal.
17
“separations”. Since the economic event we are interested in tracking is a level-one position
opening up, we do not distinguish between different types of head coach separations in our
regression models. Since the vast majority of these are forced dismissals, however, we suspect,
as the evidence in section 5 confirms, that the likelihood of separation will be negatively
correlated with performance.18 As we report in Panel B of Table 3, the overall annual rate of
separations for head coaches in our sample is 21.96%. As one would suspect given some of our
earlier figures, the figures in the table also illustrate that approximately two thirds of these
coaches are replaced by external hires.
3.4 Performance metrics and control variables
As we discuss in our introductory sections, we are particularly interested in the role of
performance measures in explaining internal and external promotions. One of the advantages of
our choice of a sports labor market is the wealth of available statistics that can be used to
construct performance metrics. We collect this statistical data for the 1970-1998 seasons from
Total Football II: The Official Encyclopedia of the National Football League (1999) and for the
1999-2001 seasons from the ESPN.com website (http://msn.espn.go.com/main.html) and from
the 2000-2002 editions of The Sporting News Pro Football Guide.
As our measure of a team’s overall performance, we construct a variable called
TEAMPERF. This variable is based on a team’s winning percentage, but with some adjustments
18 In our set of 191 head coach dismissals, only 21 are re-employed elsewhere as a head coach within one year of the point in time where we first identify their job separation. In another 21 cases a dismissed head coach accepts a level-two position elsewhere in that timeframe.
18
to assure that the variable has a stable distribution over time.19 Specifically, we construct
TEAMPERF by converting a team’s winning percentage in a given season into a percentile rank
based on the overall distribution of winning percentages in the NFL in that year.20 Thus, the
team with the highest winning percentage in a given year is assigned a TEAMPERF equal to 1,
the team with the lowest winning percentage is assigned a TEAMPERF value of 0, and the mean
of TEAMPERF each year is equal to .5.
Since we are also interested in more micro measures of performance that pertain to an
individual’s specific job responsibilities, we create an additional performance variable called
INDIVPERF. We construct this measure for offensive coordinators by ranking teams each year
by the total number of points scored during the season. We then convert this raw ranking into a
percentile rank measure, where the highest ranking team in a given season (i.e., the one with the
most points) is assigned a percentile rank of 1, and the lowest ranking team is assigned a
percentile rank of 0. For defensive coordinators, we proceed in an analogous manner, but in this
case we use points scored against the team as the measure of success. Thus, the team with the
fewest points scored against them in a given season is assigned a percentile rank of 1, and the
team with the most points scored against them receives a percentile rank of 0. The variable
INDIVPERF is then set equal to these percentile rank numbers. The advantage of this variable is
that it will, by construction, have stable distributional properties across seasons and across the
two types of positions. The mean of this variable will be .5 in each year for each type of coach,
and the maximum and minimum will always be 1 and 0 respectively.
19 Team winning percentage is defined to be [number of games won + .5 x number of games tied]/[number of games played]. 20 All percentile ranks in this paper are calculated by first constructing raw rankings (R) based on the chosen performance characteristic. The worst performer receives a raw rank of 1 and the best receives a rank equal to the number of observations in the set being ranked (N). All tied observations are assigned a raw rank equal to their median rank (e.g., if three teams are tied for 2nd place, each is assigned a raw rank of 3). We then define the percentile rank of an observation to be equal to (R-1) /(N–1).
19
We report in Panels C and D of Table 3 some summary statistics for the performance
metrics. As designed, the means of the TEAMPERF and the offensive and defensive
INDIVPERF variables are 0.50. Also, all three variables have a standard deviation of 0.30,
suggesting the three measures have similar statistical properties. Not surprisingly, the correlation
between TEAMPERF and INDIVPERF is very high (0.70-0.75). Scoring points and keeping the
opponent from scoring are very important components to winning. The correlation between the
INDIVPERF of a team’s offensive coordinator and its defensive coordinator is also positive, but
it is much smaller in magnitude (0.34). This suggests that there is some correlation between
offensive and defensive activities, but also a substantial degree of independence.21
4. Analysis of labor market promotions
4.1 External labor market promotions
We now turn to modeling outcomes where level-two coaches depart to take level-one
positions elsewhere. Our empirical approach is to run logit models where the dependent variable
assumes a value of 1 for these external promotion outcomes and a value of 0 for outcomes where
the level-two coach remains in his current position.22 We include in these models variables
related to the individual’s age and tenure with the team along with a year trend.23
21 There are many explanations for the positive correlation. If teams have high budgets, they may hire better players on both sides of the ball. Alternatively, if the offense is strong and keeps the ball for most of the game, the defense will have very little chance to be scored upon. 22 The dependent variable for all other labor market outcomes, for example dismissals, internal promotions, and lateral moves are coded as missing in the logit models of Table 4. A multinomial model where we create several different outcomes for the dependent variable is more cumbersome to report and yields results similar to those of the simpler logit models that we present in the text and tables. 23 In the regressions, year is defined as the calendar year at the start of the season minus 1970.
20
We present our basic logit models of external promotions in Table 4. In columns 1 and 2
we examine the offensive and defensive coordinators separately, while in column 3 we combine
them together. The basic performance measure in these regression is the individual performance
variable (INDIVPERF).24 The coefficients in columns 1-3 indicate that there is a very strong
relationship between an individual’s performance and the likelihood of obtaining a head
coaching position elsewhere. Since the promotion/performance sensitivities of offensive and
defensive coordinators are not significantly different, we focus primarily on the specifications
where offensive and defensive coordinator observations are pooled together.25 For the model
estimated in column 3 the estimates imply that the likelihood of a level-two coach at the 25th
percentile performance level obtaining a head coach job at a new employer is 0.98% when all
other variables are held at their means. In sharp contrast, the same coach at the 75th percentile
performance level has a 5.04% chance of obtaining this external promotion. Clearly the external
labor market looks closely at performance in their assessment of an individual’s talents. These
results are highly consistent with the principal assumption underlying the career concerns
literature of Fama (1980) and Holmström (1982). These findings are also consistent with the
recent empirical results reported by Fee and Hadlock (2002a) concerning the market for
executive talent.
In column 4 of Table 4 we add to the logit specification the team performance variable
(TEAMPERF). Interestingly, while this coefficient is positive, it is small in magnitude and
statistically insignificant (t=1.35). At the same time, the coefficient on the individual
performance variable remains positive and highly significant (t=2.85). These differences suggest
24 Given the nature of of the labor market we study and of our performance measures, we use levels of performance rather than changes in levels of performance in our regressions. This is the appropriate approach if, as NFL coaches attest, staying at the top takes at least as much talent and hard work as getting to the top in the first place. 25 As a formal test of differences in performance sensitivities, we added a term interacting position with performance to the specification of column 3 and found no statistically significant difference.
21
that the labor market relies more heavily on more micro-level performance metrics in their
assessment of an individual’s talents.26 From an incentives perspective, this suggests that level-
two coaches concerned with their external reputation will be more concerned with exhibiting
excellence in the area they oversee (offense or defense) than in overall organizational
performance (winning games). In many instances, of course, these two objectives are highly
complementary to one another.
Note that in the specification of column 4, team performance serves as a control for other
variables outside level-two coaches’ control, such as overall player quality, head coach quality,
etc. For this reason, this specification allows us to test an alternative hypothesis concerning the
observed positive external promotion/performance sensitivity; namely, that the relationship is a
function of teams trying to hire coaches who have “learned how to win” rather than a function of
inferences about individuals’ ability. Since the coefficient on individual performance remains
significant even after controlling for team performance, it appears that ability inferences are the
dominant factor in the labor market we study.
Turning to the other control variables, the most robust finding is that older coaches are
less likely to be hired as head coaches elsewhere. The insignificance on the dummy variable for
offensive vs. defensive coordinator suggests that both types of coaches have a similar likelihood
of receiving an external promotion. Finally, there is some limited evidence (at the 10% level)
that external promotions have increased in likelihood over time.
26 These results are only suggestive, as the difference in the estimated coefficients on the two performance metrics is not significant. Certainly we can say with some confidence that individual performance metrics are at least as important as team metrics in evaluating talent. We additionally experimented with replacing TEAMPERF in the specification of column 4 with a variable representing the individual’s counterparty’s performance (i.e. adding defensive INDIVPERF for the offensive coordinator and offensive INDIVPERF for the defensive coordinator.) In that specification (unreported), INDIVPERF remained positive and significant at the 1% level and the counterparty performance variable was insignificant. Additionally, the coefficient on INDIVPERF was statistically greater than the coefficient on counterparty performance at conventional levels (p value<0.05, two-tailed t-test).
22
4.2 Evidence on the internal labor market
Using the external labor market findings as a baseline, we now turn to examining the
internal promotion mechanism in our sample. For each observation, we create a dependent
variable that takes a value of 1 for an internal promotion (i.e., the individual is promoted
internally to the head coach position), and a value of 0 if the individual stays with his team and
his title is unchanged. We pool all level-two coaches together and use specifications that parallel
our external promotion analysis presented above. Our baseline results where the individual
performance metric is the only performance measure are presented in column 1 of Table 5. The
coefficient on INDIVPERF in this specification is actually negative and insignificant.27 Thus,
increased levels of individual performance in the sphere under an individual’s control do not
appear to be associated with an increased likelihood of being internally promoted to the top job.
All of the other control variables are insignificant.
From an incentives perspective, these results in column 1 of Table 5 are puzzling. They
suggest, for example, that an offensive coordinator who successfully pushes his team to score
more points will not, in fact, be rewarded with an increased likelihood of internal promotion. A
reasonable explanation for this finding is as follows. When the offense scores a high number of
points, teams tend to win. When teams win, the head coach keeps his job. Consequently, there
is no job to promote the offensive coordinator into. Analogous remarks could be made for the
defensive coordinator. This is a slot constraint type situation as discussed in the introductory
sections.
27 We are able to test the differences in performance sensitivity between internal and external promotions by including both in a multinomial logit model of job market outcomes. When we do so, we find a large and very significant (t=4.6) difference between the coefficient on INDIVPERF for internal and external promotions for the specification analogous to column 3 of Table 4 and column 1 of Table 5.
23
To further investigate, in the specification in column 2 of Table 5 we add as an additional
control variable the team’s overall performance (TEAMPERF). In this specification, the
coefficient on team performance is negative and highly significant (t=-4.01), while the
coefficient on individual performance is positive and significant (t=2.35). Thus, when firms are
winning, it does appear that the chance of an internal promotion is low. Holding constant the
winning percentage, increased performance by a given level-two coach will increase his
likelihood of an internal promotion. However, since an increased level of individual
performance tends to increase the likelihood of winning, the net effect of increasing individual
performance on the internal promotion likelihood is best reflected in the column 1 specification,
where the estimate is negative and insignificant.
The findings in columns 1 and 2 of Table 5 for the internal labor market promotions
contrast sharply with our earlier findings in Table 4 for external labor market promotions.
Clearly the net effect from raising the performance of the unit under your control on the
likelihood of getting a head coaching position differs depending on the type of head coaching job
under consideration (internal vs. external). Our results suggest that the difference in the internal
labor market arises from the relationship between performance and the likelihood of a job
opening. We will look at this more closely below where we estimate specific models of internal
job openings and a team’s decision of who to hire conditional on having an opening.
Before turning to this analysis, it is informative to examine the decision to dismiss level-
two coaches.28 To conduct this analysis, we create a dependent variable that equals 1 for level-
two coaches who were dismissed (i.e., lost their position) and did not gain new employment, and
equals 0 for level-two coaches whose position and team remain unchanged during the
28 This analysis parallels studies that examine departures of senior level executives directly below the CEO in large publicly traded corporations (e.g., Fee and Hadlock (2002b), Hayes, Oyer, and Schaefer (2002)).
24
observation-year. We report these results in columns 3 and 4 of Table 5. In column 3, where the
only performance metric is individual performance, the estimate on the performance variable is
negative and significant. If an individual is performing poorly, he is more likely to be fired.
When we add the team performance metric in column 4, we find that the coefficients on both
individual and team performance have similar estimated magnitudes which are negative and
highly significant.
These findings on individual versus team performance metrics are interesting when
viewed in contrast to our earlier findings on external labor market decisions. Our earlier results
suggest that the external labor market behaves as if micro measures of performance are more
related to an individual’s talents than are macro or team measures. However, when a franchise is
deciding to remove a level-two coach, it appears that they rely approximately equally on both
micro and macro measures. We suspect that the significance of the macro performance measure
in the firing of level-two coaches reflects the outcome of a process where, regardless of
individual performance, level-two coaches no longer “fit well” in the organization after the head
coach to which they owe their allegiance is fired.29
5. Job openings and promotions
5.1 Estimating models of job openings
The results for internal promotions in the previous section suggest that a level-two
coach’s individual performance is negatively related to the probability of a job opening. We
29 Results on the team nature of departures in the market for managerial talent are reported by Fee and Hadlock (2002b) and Hayes, Oyer, and Schaefer (2002).
25
suspect that this negative relationship arises because of the affect of team performance on the
likelihood of the head coach keeping his job. To investigate, we model here a team’s decision to
replace its head coach. These models are quite analogous to models that are estimated in the
literature on CEO turnover (e.g., Warner, Watts, and Wruck (1988) and Weisbach (1988)).
In column 1 of Table 6 we present a logit model where the dependent variable takes a
value of 1 if the head coach departs from the firm for any reason and a 0 if the head coach stays.
The coefficient on the team performance variable in this model is negative and highly significant
(t=-9.49). Holding other variables at their means, a head coach whose team is at the 25th
percentile level of performance has a 30.90% chance of losing his position, while the
corresponding figure for a team at the 75th percentile performance level is only 7.73%.30 Clearly
head coaches are held very responsible for team performance, and consequently promotion
opportunities for subordinates are likely to be negatively related to team performance. Since
individual performance is strongly correlated with team performance, promotion opportunities
for level-two coaches who exhibit strong individual performance should be quite limited.
The internal promotion opportunities for a level-two coach will depend not only on a
team’s decision to remove the head coach, but also on its decision to choose an internal
replacement. It is well known that firms tend to choose outside CEOs when they are performing
particularly poorly (e.g., Parrino (1997)). If similar behavior occurs in the coaching market, then
strong team performance may increase the likelihood that the team promotes from within when
an opening arises.31
30 Note that some job separations in these models are likely to be voluntary retirements rather than dismissals. This could explain why the implied probability of departure is still non-trivial for strong performing head coaches. This would also explain why the estimated coefficients on the age and tenure variables are positive, as these variables are likely to proxy for voluntarily departures and retirements. 31 In a similar vein, Blackwell, Brickley, and Weisbach (1993) show that, conditional on a job opening at the head of a division, banks tend to be more likely to promote insiders to the division head position when the division is performing well.
26
To investigate, in column 2 (column 3) we run our logit model on a dependent variable
that takes a value of 1 when firms replace their head coach and hire an outside replacement
(inside replacement), and a value of 0 when the head coach remains with the team. The
estimated coefficient on team performance is negative and significant in columns 2 and 3, but the
magnitude of the estimated coefficient is substantially larger in the outside replacement model
(column 2) than it is in the insider replacement model (column 3).32 Thus, it does appear that
poorer performance leads teams towards more outside hiring when the head coach is dismissed.
However, since head coaches generally are dismissed following poor performance, the overall
relationship between team performance and the likelihood of an insider being promoted in any
given year is still negative.
The evidence here clearly suggests that the weak relationship between individual
performance and internal promotions is highly influenced by the endogenous process governing
job openings. Teams that replace the head coach and promote an insider tend to be exhibiting
poorer than average performance. Since strong individual performance tends to lead to strong
team performance, the better performing level-two coaches are unlikely to have an internal job to
be promoted into. This evidence, along with our earlier findings indicating substantial
movement of level-two coaches across teams, suggests that internal slot constraints are often
binding constraints in the coaching labor market
5.2 Choosing amongst internal candidates
32 Formal tests for differences in the team performance coefficient in the corresponding multinomial logit model reveals that this difference is significant at the 1% confidence level.
27
Certainly performance must count for something in the internal promotion process. To
identify whether this is the case, our preceding findings suggest that we need to condition our
analysis on the existence of a job opening. Thus, we investigate here the role of an individual’s
performance on the likelihood of a promotion conditional on the team promoting an insider. To
conduct this analysis, we examine every case in the sample where the firm had both an offensive
and defensive coordinator and where one of these individuals was promoted to the head coach
position. We then use McFadden’s (1973) conditional logit methodology to estimate a model
predicting the team’s choice of one of these two internal candidates over the other. The key
independent variable of interest here is the individual performance measure.
The results from a very simple conditional logit model where the only independent
variable is individual performance are reported in column 1 of Table 7. Even though the sample
is fairly small (24 promotions in sample of 2 x 24 = 48 individuals), the estimated coefficient on
the internal performance measure has the expected positive sign, and it is significant at the 5%
level (t=2.07)33. To gauge the magnitude of the effect, suppose that one of the level-two coaches
exhibits performance at the 50th percentile level. The estimates in column 1 imply that if the
other level-two coach raises his performance from the 25th percentile level to the 75th percentile
level, his likelihood of being chosen for the internal promotion increases from 28.7% up to
71.3%. It appears that when firms promote insiders, they rely heavily on individual performance
metrics in evaluating the candidates.
To check the robustness of these findings, we include in column 2 of Table 7 several
control variables that may affect a team’s promotion choice including age, tenure, and a dummy
variable that distinguishes between offensive and defensive coordinators. As the estimates
33 The number of internal promotions in this regression is less than the total number of internal promotions in the sample because a team must have both a designated offensive coordinator and a designated defensive coordinator to be included in the conditional logit.
28
reveal, none of these added control variables is significant at conventional levels. The
performance variable remains positive, although its significance level falls slightly to the 10%
level (t=1.70). The estimated economic magnitude of the role of performance on the choice of
who to promote remains large in this specification.
Putting together the results from Tables 6 and 7, we can now make some sense of our
earlier finding of no positive relationship between promotions and the probability of an internal
promotion. Using our estimates in column 3 of Table 6, when a team’s performance increases
from the 25th percentile level to the 75th percentile level, the probability that one of the two level-
two internal coaches gets promoted during the subsequent year decreases from 10.33% down to
4.02%. However, when individual performance increases by the same amount in percentile
terms, the probability of a given level-two coach getting the job when it is offered to an insider
increases from 28.7% up to 71.3%. In light of these estimated magnitudes, along with the high
correlation between team performance and individual performance, it is straightforward to
observe how the relationship between promotions and individual performance could be fairly
flat, or even negative.
6. Lateral movements and relationships
6.1 Lateral moves
Most of our focus up to this point has been on labor market events that can clearly be
classified as promotions. However, as our early summary statistics on coaching backgrounds
revealed, there is a great deal of lateral movement across teams where level-two coaches switch
29
employers without changing titles. This type of job-hopping is consistent with matching models
where coaches search for employers where they fit well (e.g., Jovanovic (1979)). We suspect,
however, that issues of matching and fit in the coaching market are often driven by the closeness
of the relationship between head coaches and their subordinates. In particular, a level-two coach
may become a poor match for a firm after the head coach departs because of the loss of the
relationship-specific human capital that was formed between the head coach and his assistants.
To investigate these issues, we estimate logit models predicting lateral moves for level-
two coaches. The dependent variable in these models takes a value of 1 if a level-two coach
moves laterally to a new employer, and a zero if he keeps his position. In column 1 of Table 8
we report results where the only performance variable is the individual performance metric. The
estimated coefficient on individual performance in this specification is negative and significant
(t=-4.28), indicating that coaches tend to move laterally following bouts of poor individual
performance. In column 2 of Table 8 we add the team performance variable to the model. In
this specification, the estimated coefficient on the individual performance variable becomes
small in magnitude and statistically insignificant, while the coefficient on the team performance
variable is negative and highly significant (t=-3.86). These estimates suggest that it is poor team
performance rather than poor individual performance that drives coaches to move laterally. This
is what we would expect if poor team performance leads to an increased likelihood of a head
coach dismissal, which in turn may cause level-two coaches to look to make a lateral move.
It is interesting to compare these results in Table 8 on lateral moves for level-two coaches
to our earlier Table 5 findings on dismissals of level-two coaches. It is clear from our estimates
that poor team performance increases the likelihood of both a dismissal and a lateral move.
However, after controlling for team performance, it appears that individual performance does not
30
independently affect the likelihood of a lateral move, but it certainly affects the likelihood of an
outright dismissal (see Table 5). These findings suggest that separating from an employer is
primarily related to the current employer’s team performance, while job prospects elsewhere are
primarily related to individual performance.
This suspicion is borne out in the findings in columns 3 and 4 of Table 8. Here we ask
the question of whether a level-two coach gains new-employment as a level-two coach at a new
team conditional on separating from his old team (and not obtaining a head-coach position). In
these logit models, the coefficient on individual performance is positive and highly significant,
while the coefficient on team performance is negative and insignificant. Consistent with our
earlier findings on external promotions, these findings suggest that the external labor market
relies primarily on individual performance in its assessment of talent.
6.2 Relationships in hiring and firing
The preceding findings suggest that many job separations of level-two coaches are driven
by the departure of the head coach. To investigate directly, we examine the correlation in
departure rates of level-two coaches and head coaches. As we report in Panel A of Table 9, we
observe that 73.68% of all level-two coaches depart when their head coach departs. This figure
increases to 86.21% when the replacement head coach is an outsider. This suggests that the head
coaching staff is viewed as a team, and they typically get fired as a group. This is what we
would expect if the value of a level-two coach to his employer depends on the closeness of his
relationship with the head coach.
31
If relationship specific capital affects the value of a coach to a team, we would expect it
to affect hiring decisions as well as firing decisions. To examine whether this is the case, we
identify in our sample every case where an outsider is hired as a head coach. We then
investigate the identities of the level-two coaches that the head coach appoints to serve under
him. We find that, when a newly appointed head coach appoints a new offensive or defensive
coordinator, 19.66% of the time that individual comes from the same prior team as the head
coach (see Panel B of Table 9). If the head coach were hiring randomly from the pool of all NFL
teams, we would predict a figure here of 3.51%. A formal statistical test reveals that the
observed figure differs from the predicted figure at high levels of significance (p-value < 0.01).
Thus, the data strongly indicate that coaches have a substantial propensity to hire coaches with
whom they have established a relationship, presumably to take advantage of relationship specific
human capital.
7. Conclusion
In this study we examine labor market outcomes for high-level coaches employed by
National Football League teams from 1970 to 2001. Our analysis indicates that there are
significant differences in the mechanism governing internal labor markets and external labor
markets. In particular, we find that objective measures of a level-two coach’s individual
performance are significantly related to the likelihood that the coach obtains a promotion from
the outside labor market in the form of a more prestigious position (i.e., a head coach/level-one
position). In contrast, individual performance measures appear to be unrelated to the overall
likelihood of an internal promotion.
32
We hypothesize that this difference between internal and external labor markets arises
from the process governing job openings within the internal hierarchy, and we find evidence
consistent with this hypothesis. Our evidence indicates that the likelihood of a job opening at the
top of the coaching hierarchy is negatively related to team performance. Since there is a strong
causal link between individual performance and team performance, an increase in individual
performance has a negative effect on the likelihood of a job opening appearing. Conditional on
an internal opening occurring, we do find that increases in individual performance increase the
probability of being promoted. It appears that these two effects roughly cancel out, resulting in
an overall flat relationship between individual performance and the unconditional likelihood of
an internal promotion.
Our evidence on external labor market transactions indicates that prospective employers
look carefully at objective measures of individual performance in assessing a level-two coach’s
ability, while we find no strong evidence that outsiders also use team performance in making
these ability inferences. This finding may have interesting implications in other labor markets
where firms have some control over the information metrics available to outsiders. For example,
firms often have some organizational design and accounting choices that affect the information
flow to outsiders. These choices may affect both the incentives of their managerial employees
and the behavior of outsiders looking to hire the best of these employees.
The findings we present also suggest that relationships matter a great deal in this labor
market. In particular, we find that coaches are often dismissed as a group. When the head coach
is fired, his subordinates are very likely to also separate from the franchise. In addition, we find
evidence that coaches are hired as a group. In particular, when a new head coach is hired, he
tends to bring along subordinates with whom he has had a past relationship. This evidence on
33
hiring and firing suggests that the value of an individual to an employer depends on the identity
of the entire set of individuals who work together, and the value of this group as a whole depends
on the closeness of their accumulated stock of relationship-specific human capital.
While the results we present are drawn from a specific labor market, we believe that
some general lessons can be drawn from our empirical analysis. First, our evidence suggests that
internal promotion incentives to increase performance may not be very strong at the top of an
organizational hierarchy. Since there are very few top positions to go around, and since the
internal prospects of individuals at the top are closely related, it is quite rare for strong individual
performance to lead to an internal promotion. Thus, while some have argued that tournament
type promotion incentives in organizations are very strong, our evidence here casts doubt on the
generality of this assertion. Certainly more work, both theoretical and empirical, is needed to
fully understand how job constraints and the endogenous process governing job openings affect
careers and incentives in organizational hierarchies.
A second lesson that can be drawn from our evidence is that there may be some important
differences between an individual’s overall incentive to raise his team’s performance and his
incentive to raise his individual performance. Since individuals tend to be fired as a group, there
is a strong individual incentive to raise overall team performance to avoid the unhappy outcome
where everyone gets the axe. At the same time, if one can somehow lower team performance
while exhibiting strong individual performance, there is a chance of getting the coveted internal
promotion. Finally, independent of the level of team performance, it is clear that strong
individual performance enhances external labor market opportunities. Sorting out the relative
strength of the incentives to raise these performance metrics is an interesting question for future
research.
34
A third lesson that can be drawn from our analysis is that external labor markets can be
quite active and can serve as important incentive devices. While professional football coaches
switch employers more than most, there is ample evidence of substantial movement across
employers in other labor markets, notably the market for managerial talent in large corporations
(e.g., Fee and Hadlock (2002a, 2002b)). In organizations where external labor markets are
active, our evidence suggests that firms should carefully consider the information concerning
individual performance that outsiders can observe.
Finally, we note here that our results suggest that there may be some interesting
relationships between how jobs are assigned, both internally and externally, and other
dimensions of the employment relationship, in particular compensation policies. For example,
following the thinking of Gibbons and Murphy (1992), it may be that when labor markets reward
certain outcomes, for example high levels of individual performance, these outcomes are less
heavily emphasized in explicit compensation arrangements. These types of possibilities suggest
that understanding the role of internal and external labor markets may be a particularly
interesting topic for future research.
35
References
Baker, George P., Michael Gibbs, and Bengt Holmström, 1994a, The internal economics of the firm: Evidence from personnel data, Quarterly Journal of Economics 109, 881-919.
Baker, George P., Michael Gibbs, and Bengt Holmström, 1994b, The wage policy of a firm,
Quarterly Journal of Economics 109, 921-955. Baker, George P., Michael C. Jensen, and Kevin J. Murphy, 1988, Compensation and incentives:
Practice vs. theory, Journal of Finance 43, 593-616. Blackwell, David W., James A. Brickley, and Michael S. Weisbach, 1994, Accounting information
and internal performance evaluation: evidence from Texas banks, Journal of Accounting and Economics 17, 331-358.
Bognanno, Michael L., 2001, Corporate tournaments, Journal of Labor Economics 19, 290-315. Carroll, Bob, Michael Gershman, David Neft, and John Thorn (eds.), 1999, Total Football II: The
Official Encyclopedia of the National Football League, Harper Collins: New York. Carter, Craig (ed.), 2000, 2001, 2002, The Sporting News Pro Football Guide, Sporting News
Publishing Co.: St. Louis. Chan, William, 1996, External recruitment vs. internal promotions, Journal of Labor Economics 14,
555-570. Fama, Eugene F., 1980, Agency problems and the theory of the firm, Journal of Political Economy
88, 288-307. Fee, C. Edward and Charles J. Hadlock, 2002a, Raids, rewards, and reputations in the market for
managerial talent, Review of Financial Studies (forthcoming). Fee, C. Edward and Charles J. Hadlock, 2002b, Management turnover across the corporate
hierarchy, Working Paper, Michigan State University. Gibbons, Robert and Kevin J. Murphy, 1992, Optimal incentive contracts in the presence of career
concerns: theory and evidence, Journal of Political Economy 100, 468-505. Gibbons, Robert and Michael Waldman, 1999a, A theory of wage and promotion dynamics inside
firms, Quarterly Journal of Economics 114, 1321-1358. Gibbons, Robert and Michael Waldman, 1999b, Careers in organizations: Theory and evidence, in
Handbook of Labor Economics - Volume 3, O. Ashenfelter and D. Card (eds.), North Holland. Gibbons, Robert and Michael Waldman, 2002, Enriching a theory of wage and promotion dynamics
inside firms, Working Paper, MIT.
36
Gibbs, Michael, 1995, Incentive compensation in a corporate hierarchy, Journal of Accounting and
Economics 19, 247-277. Hayes, Rachel M., Paul Oyer, and Scott Schaefer, 2002, Stability of top management teams,
Working paper, Stanford University. Holmström, Bengt, 1982, Managerial incentives schemes- a dynamic perspective, in Essays in
Economics and Management in Honor of Lars Wahlbeck, Swenska Handelshögkolan: Helsinki. Jovanovic, Boyan, 1979, Job matching and the theory of turnover, Journal of Political Economy 87,
972-990. Lamb, Kevin, 1999, The evolution of strategy, in Total Football II: The Official Encyclopedia of the
National Football League, B. Carroll, M. Gershman, D. Neft, J. Thorn (eds.), Harper Collins: New York.
Lazear, Edward P., 1986, Raids and offer matching, in Research in Labor Economics, Volume 8
(Part A), pp. 141-65, JAI Press: Greenwich, CT. Lazear, Edward P., 1992, The job as a concept, in Performance Measurement, Evaluations, and
Incentives, W. Bruns (ed.), Harvard University Press: Boston.
Lazear, Edward P. and Sherwin Rosen, 1981, Rank-order tournaments as optimum labor contracts, Journal of Political Economy 89, 841-864.
MacDonald, Glenn M. and James R. Markusen, 1985, A rehabilitation of absolute advantage,
Journal of Political Economy 93, 277-297. Main, Brian G. M., Charles A. O’Reilly III, and James Wade, 1993, Top executive pay:
Tournament or teamwork?, Journal of Labor Economics 11, 606-628. McFadden, Daniel, 1973, Conditional logit analysis of qualitative choice behavior, in Frontiers in
Econometrics, P Zarembka (ed.), Academic Press: New York. Medoff, James and Katharine Abraham, 1980, Experience, performance, and earnings, Quarterly
Journal of Economics 95, 703-736. Murphy, Kevin, 1985, Corporate performance and managerial remuneration: An empirical analysis,
Journal of Accounting and Economics 7, 11-42. Parrino, Robert, 1997, CEO turnover and outside succession a cross-sectional analysis, Journal of
Financial Economics 46, 165-97. Rosen, Sherwin, 1982, Authority, control, and the distribution of earnings, Bell Journal of
Economics 13-2, 311-323.
37
Rosen, Sherwin, 1986, Prizes and incentives in elimination tournaments, American Economic
Review 76, 701-715. Rosen, Sherwin and Allen R. Sanderson, 2000, Labor markets in professional sports, Working
Paper, University of Chicago. Waldman, Michael, 1984a, Worker allocation, hierarchies, and the wage distribution, Review of
Economic Studies 51, 95-109. Waldman, Michael, 1984b, Job assignments, signaling, and efficiency, RAND Journal of Economics
15, 255-267. Warner, Jerold, Ross L. Watts, and Karen H. Wruck, 1988, Stock prices and top management
changes, Journal of Financial Economics 20, 461-492. Weisbach, Michael S., 1988, Outside directors and CEO turnover, Journal of Financial Economics
20, 431-460. Wolf, Ron and Paul Attner, 1999, The Packer Way: Nine Stepping Stones to Building a Winning
Organization, St. Martin’s Griffin: New York.
38
Table 1: Summary Statistics Sample Period 1970-2001 Total Number of Franchises 31 Number of Franchises in 1970 26 Number of Franchises in 1980 28 Number of Franchises in 1990 28 Number of Franchises in 2001 31 Number of Franchise-Years 901 Number of Individuals 369 Number of Person-Years 2284 Head Coach Observations 901 Mean Age Head Coach 49.4 Mean Yrs. Pro Experience for Head Coach 13.9 Mean Yrs. with Team for Head Coach 4.4 Mean Yrs. in Position for Head Coach 3.4 Offensive Coordinator Observations 607 Mean Age Offensive Coordinator 48.0 Mean Yrs. Pro Experience for Offensive Coordinator 11.8 Mean Yrs. with Team for Offensive Coordinator 2.5 Mean Yrs. in Position for Offensive Coordinator 1.5 Number of Teams with an Offensive Coordinator in 1970 7 Number of Teams with an Offensive Coordinator in 1980 12 Number of Teams with an Offensive Coordinator in 1990 22 Number of Teams with an Offensive Coordinator in 2001 30 Defensive Coordinator Observations 776 Mean Age Defensive Coordinator 47.9 Mean Yrs. Pro Experience for Defensive Coordinator 12.6 Mean Yrs. with Team for Defensive Coordinator 3.2 Mean Yrs. in Position for Defensive Coordinator 2.0 Number of Teams with an Defensive Coordinator in 1970 11 Number of Teams with an Defensive Coordinator in 1980 26 Number of Teams with an Defensive Coordinator in 1990 28 Number of Teams with an Defensive Coordinator in 2001 31 Note: The sample of coaches is drawn from all National Football League teams from 1970 to 2001. Team histories are collected from Total Football II: The Official Encyclopedia of the National Football League for the 1970-1998 seasons and from the ESPN.com website (http://msn.espn.go.com/main.html) and the 2000-2002 editions of The Sporting News Pro Football Guide for the 1999-2001 seasons. For each team, we record the identity of the head coach, the offensive coordinator, and the defensive coordinator at the start of each annual season. The primary source of data on coaches’ identities and backgrounds are the teams’ annual media guides archived at the Pro Football Hall of Fame in Canton, Ohio. We supplemented this information when necessary with press clippings in the Hall of Fame library and with news searches on the Dow Jones Interactive electronic database.
39
Table 2: Typical NFL Coaching Career Paths
Panel A: Previous Team Worked For Head Coach Offensive Coord. Defensive Coord. Same 30.00% 29.27% 31.87% Different NFL 51.82% 64.23% 62.55% Different Pro: Non-NFL 4.55% 1.63% 2.39% College 13.64% 4.88% 3.19% Panel B: Previous Position for Internal Hires Head Coach Offensive Coord. Defensive Coord. Head Coach NA 1.41% 0.00% Offensive Coordinator 21.88% NA 0.00% Defensive Coordinator 32.81% 1.41% NA Level-Three Offensive Coach 18.75% 88.73% 0.00% Level-Three Defensive Coach 15.63% 2.82% 95.00% Other 10.94% 4.23% 5.00% Panel C: Previous Position for External Hires Head Coach Offensive Coord. Defensive Coord. Head Coach 55.26% 15.48% 13.33% Offensive Coordinator 17.11% 29.76% 0.00% Defensive Coordinator 15.79% 0.00% 39.39% Level-Three Offensive Coach 8.55% 51.19% 2.42% Level-Three Defensive Coach 1.32% 0.00% 41.82% Other 1.97% 1.79% 3.03% Note: The table describes the previous positions held by the individuals in our sample, with each person/team/position match treated as one observation. The sample of coaches is drawn from all National Football League teams from 1970 to 2001. Team histories are collected from Total Football II: The Official Encyclopedia of the National Football League for the 1970-1998 seasons and from the ESPN.com website (http://msn.espn.go.com/main.html) and the 2000-2002 editions of The Sporting News Pro Football Guide for the 1999-2001 seasons. For each team, we record the identity of the head coach, the offensive coordinator, and the defensive coordinator at the start of each annual season. The primary source of data on coaches’ identities and backgrounds are the teams’ annual media guides archived at the Pro Football Hall of Fame in Canton, Ohio. We refer to any non-coordinator assistant coaching position with offensive (defensive) responsibilities as a level-three offensive (defensive) position.
40
Table 3: Summary Statistics on Job Outcomes and Performance Variables
Panel A: Job Outcomes for Offensive and Defensive Coordinators (Level-Two Coaches) Offensive Coordinators Defensive Coordinators No change 66.26% 69.53% Dismissal 18.76% 16.11% Lateral Movement 9.00% 8.99% External Promotion 3.40% 2.68% Internal Promotion 2.57% 2.68% Panel B: Job Outcomes for Head Coaches (Level-One Coaches) No Change 78.05% Separation / Replaced Internally 6.67% Separation / Replaced Externally 15.29% Panel C: Means and Standard Deviations of Performance Metrics Mean Standard Deviation TEAMPERF 0.50 0.30 Offensive INDIVPERF 0.50 0.30 Defensive INDIVPERF 0.50 0.30 Panel D: Correlation Matrix of Performance Metrics
TEAMPERF Offensive
INDIVPERF Defensive
INDIVPERF TEAMPERF 1.00 Offensive INDIVPERF 0.74 1.00 Defensive INDIVPERF 0.70 0.34 1.00 Note: For Panels A and B, we treat each person-year as one observation and for Panels C and D, we treat each team-year as one observation. The sample of coaches is drawn from all National Football League teams from 1970 to 2001. Team histories and performance statistics are collected from Total Football II: The Official Encyclopedia of the National Football League for the 1970-1998 seasons and from the ESPN.com website (http://msn.espn.go.com/main.html) and the 2000-2002 editions of The Sporting News Pro Football Guide for the 1999-2001 seasons. For each team, we record the identity of the head coach, the offensive coordinator, and the defensive coordinator at the start of each annual season. The primary source of data on coaches’ identities and backgrounds are the teams’ annual media guides archived at the Pro Football Hall of Fame in Canton, Ohio. If an individual remains with the same team in the same position at the start of the subsequent season, we code this as “No Change”. “Internal Promotions” are cases where an offensive or defensive coordinator moves to a head coaching position with the same team in the subsequent season. “External Promotions” are cases when an offensive or defensive coordinator moves to a head coaching position within one year of the start of the season where we identify a separation; “Lateral Moves” refer to cases where they move to an offensive or defensive coordinator at a different team. All other separations are referred to as “Dismissals”. For head coaches, we categorize all cases where they leave their team as “Separations”. “TEAMPERF” is a team’s annual percentile rank (as defined below) in total winning percentage ([Number of wins + 0.5 * Number of Ties]/[Total Games Played]). “INDIVPERF” for offensive (defensive) coordinators is the team’s annual percentile rank in points scored (points scored against). Percentile rank, defined as [Raw Rank –1]/[N – 1], normalizes the performance distribution into a 0 to 1 scale with 0 being the worst, 1 being the best, and 0.5 being the mean.
41
Table 4: Performance and External Promotions for Level-Two Coaches
Specification (1) (2) (3) (4) Coaches Included OC DC OC and DC OC and DC INDIVPERF 3.714
(1.063)*** 2.919
(1.01)*** 3.365
(0.731)*** 2.590
(0.910)*** TEAMPERF 1.186
(0.881) Age -0.081
(0.031)*** -0.038 (0.032)
-0.062 (0.022)***
-0.061 (0.022)***
Years with Team 0.050
(0.086) 0.026
(0.064) 0.033
(0.051) 0.027
(0.052) Year 0.021
(0.027) 0.055
(0.029)* 0.037
(0.020)* 0.036
(0.020)* OC vs. DC 0.514
(0.319) 0.507
(0.320) Constant -2.084
(1.646) -4.489
(1.692)*** -3.437
(1.189)*** -3.718
(1.209)*** Nobs. 383 538 921 921 Pseudo R2 14.47% 8.64% 11.96% 12.94% Predicted Probabilities vs. Performance (INDIVPERF) 75th Percentile 6.26% 4.25% 5.04% 3.96% 25th Percentile 1.03% 1.02% 0.98% 1.12% Difference 5.23% 3.23% 4.06% 2.84%
Note: The dependent variable in the reported logit specifications is a dummy variable set equal to 1 for external promotions and 0 for no change observations, both as defined in Table 3. The dependant variable is set to missing for all other observations. The sample of coaches is drawn from all National Football League teams from 1970 to 2001. Team histories and performance statistics are collected from Total Football II: The Official Encyclopedia of the National Football League for the 1970-1998 seasons and from the ESPN.com website (http://msn.espn.go.com/main.html) and the 2000-2002 editions of The Sporting News Pro Football Guide for the 1999-2001 seasons. For each team, we record the identity of the head coach, the offensive coordinator, and the defensive coordinator at the start of each annual season. The primary source of data on coaches’ identities and backgrounds are the teams’ annual media guides archived at the Pro Football Hall of Fame in Canton, Ohio. “TEAMPERF” is a team’s annual percentile rank (as defined below) in total winning percentage ([Number of wins + 0.5 * Number of Ties]/[Total Games Played]). “INDIVPERF” for offensive (defensive) coordinators is the team’s annual percentile rank in points scored (points scored against). Percentile rank, defined as [Raw Rank –1]/[N – 1], normalizes the performance distribution into a 0 to 1 scale with 0 being the worst, 1 being the best, and 0.5 being the mean. Year is equal to the calendar year at the start of the season minus 1970. Predicted probabilities are derived from the logit models by varying INDIVPERF, while maintaining all other variables at their means. Reported are coefficient estimates with estimated standard errors in parentheses. *** (**, *) Significant at the 1% (5%, 10%) confidence level using a two-tailed t-test.
42
Table 5: Internal Incentives for Level-Two Coaches
Specification (1) (2) (3) (4) Dependent Variable Promotion Promotion Dismissal Dismissal INDIVPERF -0.410
(0.610) 1.973
(0.841)** -2.906
(0.292)*** -1.859
(0.392)*** TEAMPERF -3.599
(0.897)*** -1.497
(0.387)*** Age -0.019
(0.025) -0.021 (0.26)
0.038 (0.011)***
0.039 (0.011)***
Years with Team 0.019
(0.050) 0.038
(0.049) 0.012
(0.022) 0.021
(0.022) Year -0.001
(0.021) 0.001
(0.022) -0.011 (0.010)
-0.011 (0.010)
OC vs. DC 0.039
(0.359) 0.098
(0.363) 0.445
(0.157)*** 0.450
(0.158)*** Constant -2.182
(1.211) -1.771 (1.254)
-1.893 (0.548)***
-1.755 (0.556)**
Nobs. 910 910 1120 1120 Pseudo R2 0.36% 6.27% 12.05% 13.36% Predicted Probabilities vs. Performance (INDIVPERF) 75th Percentile 3.38% 5.09% 10.04% 12.66% 25th Percentile 4.12% 1.96% 32.32% 26.86% Difference -0.74% 3.13% -22.28% -14.20%
Note: The dependent variable in the logit specifications reported in Columns (1) and (2) is a dummy variable set equal to 1 for internal promotions and 0 for no change observations, both as defined in Table 3. For the models reported in Columns (3) and (4) it is set equal to 1 for dismissals and 0 for no change observations, also as defined in Table 3. The dependent variable is set to missing for all other observations. The sample of coaches is drawn from all National Football League teams from 1970 to 2001. Team histories and performance statistics are collected from Total Football II: The Official Encyclopedia of the National Football League for the 1970-1998 seasons and from the ESPN.com website (http://msn.espn.go.com/main.html) and the 2000-2002 editions of The Sporting News Pro Football Guide for the 1999-2001 seasons. For each team, we record the identity of the head coach, the offensive coordinator, and the defensive coordinator at the start of each annual season. The primary source of data on coaches’ identities and backgrounds are the teams’ annual media guides archived at the Pro Football Hall of Fame in Canton, Ohio. “TEAMPERF” is a team’s annual percentile rank (as defined below) in total winning percentage ([Number of wins + 0.5 * Number of Ties]/[Total Games Played]). “INDIVPERF” for offensive (defensive) coordinators is the team’s annual percentile rank in points scored (points scored against). Percentile rank, defined as [Raw Rank –1]/[N – 1], normalizes the performance distribution into a 0 to 1 scale with 0 being the worst, 1 being the best, and 0.5 being the mean. Year is equal to the calendar year at the start of the season minus 1970. Predicted probabilities are derived from the logit models by varying INDIVPERF, while maintaining all other variables at their means. Reported are coefficient estimates with estimated standard errors in parentheses. *** (**, *) Significant at the 1% (5%, 10%) confidence level using a two-tailed t-test.
43
Table 6: Head Coaches’ Separations
Specification (1) (2) (3) Dependent Variable Separation
vs. No Change Separation/ Outside
Replacement vs. No Change
Separation/ Internal Replacement
vs. No Change TEAMPERF -3.350
(0.353)*** -4.13
(0.449)*** -2.024
(0.525)*** Age 0.038
(0.014)*** 0.032
(0.016)* 0.046
(0.021)** Years with Team 0.050
(0.019)*** 0.061
(0.022)*** 0.034
(0.028) Year -0.011
(0.010) -0.006 (0.012)
-0.022 (0.016)
Constant -1.804
(0.671)*** -1.761
(0.794)** -3.627
(1.044)*** Nobs. 870 812 737 Pseudo R2 13.82% 17.26% 5.97% Predicted Probabilities vs. Performance (TEAMPERF) 75th Percentile 7.73% 3.97% 4.02% 25th Percentile 30.90% 24.58% 10.33% Difference -23.17% -20.61% -6.31% Note: The dependent variable in all above specifications equals 0 for no change observations, as defined in Table 3. The dependent variable takes on the value of 1 for various subsets of separations, also defined in Table 3. In all other cases, the dependent variable is set to missing. The sample of coaches is drawn from all National Football League teams from 1970 to 2001. Team histories and performance statistics are collected from Total Football II: The Official Encyclopedia of the National Football League for the 1970-1998 seasons and from the ESPN.com website (http://msn.espn.go.com/main.html) and the 2000-2002 editions of The Sporting News Pro Football Guide for the 1999-2001 seasons. For each team, we record the identity of the head coach, the offensive coordinator, and the defensive coordinator at the start of each annual season. The primary source of data on coaches’ identities and backgrounds are the teams’ annual media guides archived at the Pro Football Hall of Fame in Canton, Ohio. “TEAMPERF” is a team’s annual percentile rank (as defined below) in total winning percentage ([Number of wins + 0.5 * Number of Ties]/[Total Games Played]). Percentile rank, defined as [Raw Rank –1]/[N – 1], normalizes the performance distribution into a 0 to 1 scale with 0 being the worst, 1 being the best, and 0.5 being the mean. Year is equal to the calendar year at the start of the season minus 1970. Predicted probabilities are derived from the logit models by varying TEAMPERF and maintaining all other variables at their means. Reported are coefficient estimates with estimated standard errors in parentheses. *** (**, *) Significant at the 1% (5%, 10%) confidence level using a two-tailed t-test.
44
Table 7: Conditional Logit Models of Internal Promotion
Specification (1) (2) INDIVPERF 3.640
(1.761)** 3.499
(2.064)* Age -0.080
(0.064) Years with Team -0.117
(0.143) OC vs. DC 0.532
(0.543) Nobs. 48 48 Pseudo R2 18.37% 28.95% Predicted Probabilities vs. Performance (INDIVPERF) 75th Percentile 71.30% 54.52% 25th Percentile 28.70% 17.25% Difference 42.60% 37.27%
Note: Reported are conditional logit (McFadden (1972)) models predicting promotion probabilities for offensive and defensive coordinators conditional on (1) the team having both slots filled at the start of one year and (2) one of the individuals serving as the team’s head coach at the start of the next year. The dependent variable in the reported logit specifications is a dummy variable set equal to 1 for internal promotions as defined in Table 3. The variable equals 0 in all other cases where an individual has the title of offensive or defensive coordinator at the start of the season. “INDIVPERF” for offensive (defensive) coordinators is the team’s annual percentile rank in points scored (points scored against). Percentile rank, defined as [Raw Rank –1]/[N – 1], normalizes the performance distribution into a 0 to 1 scale with 0 being the worst, 1 being the best, and 0.5 being the mean. The sample of coaches is drawn from all National Football League teams from 1970 to 2001. Team histories and performance statistics are collected from Total Football II: The Official Encyclopedia of the National Football League for the 1970-1998 seasons and from the ESPN.com website (http://msn.espn.go.com/main.html) and the 2000-2002 editions of The Sporting News Pro Football Guide for the 1999-2001 seasons. For each team, we record the identity of the head coach, the offensive coordinator, and the defensive coordinator at the start of each annual season. The primary source of data on coaches’ identities and backgrounds are the teams’ annual media guides archived at the Pro Football Hall of Fame in Canton, Ohio. Predicted probabilities are derived by varying INDIVPERF for one of the two coaches, holding INDIVPERF at the 50% percentile for the other coach, and maintaining all other variables at their means. Reported are coefficient estimates with estimated standard errors in parentheses. Reported are coefficient estimates with estimated standard errors in parentheses. *** (**, *) Significant at the 1% (5%, 10%) confidence level using a two-tailed t-test.
45
Table 8: Lateral Movements for Level-Two Coaches
Specification (1) (2) (3) (4) Dependent Variable Lateral vs.
No Change Lateral vs. No Change
Lateral vs. Dismissal
Lateral vs. Dismissal
INDIVPERF -1.527
(0.357)*** -0.231 (0.487)
1.434 (0.415)***
1.792 (0.581)***
TEAMPERF -1.921
(0.498)*** -0.525
(0.592) Age 0.031
(0.014)** 0.032
(0.05)** -0.125 (0.016)
-0.013 (0.164)
Years with Team 0.061
(0.026)*** 0.072
(0.026)*** 0.060
(0.034)* 0.065
(0.034)* Year 0.028
(0.013)** 0.028
(0.013)*** 0.029
(0.014)** 0.030
(0.014)** OC vs. DC 0.149
(0.203) 0.156
(0.204) -0.245 (0.234)
-0.229 (0.235)
Constant -3.527
(0.729)*** -3.318 (0.740)
-1.246 (0.804)
-1.201 (0.808)
Nobs. 995 995 363 363 Pseudo R2 4.77% 6.89% 4.89% 5.06% Predicted Probabilities vs. Performance (INDIVPERF) 75th Percentile 8.39% 11.03% 45.69% 47.35% 25th Percentile 16.43% 12.22% 29.12% 26.85% Difference -8.04% -1.18% 16.58% 20.50%
Note: The dependent variable in the logit specifications reported in Columns (1) and (2) is a dummy variable set equal to 1 for lateral moves and 0 for no change observations, both as defined in Table 3. For the models reported in Columns (3) and (4) it is set equal to 1 for lateral moves and 0 for dismissals, also as defined in Table 3. The dependent variable is set to missing for all other observations. The sample of coaches is drawn from all National Football League teams from 1970 to 2001. Team histories and performance statistics are collected from Total Football II: The Official Encyclopedia of the National Football League for the 1970-1998 seasons and from the ESPN.com website (http://msn.espn.go.com/main.html) and the 2000-2002 editions of The Sporting News Pro Football Guide for the 1999-2001 seasons. For each team, we record the identity of the head coach, the offensive coordinator, and the defensive coordinator at the start of each annual season. The primary source of data on coaches’ identities and backgrounds are the teams’ annual media guides archived at the Pro Football Hall of Fame in Canton, Ohio. “TEAMPERF” is a team’s annual percentile rank (as defined below) in total winning percentage ([Number of wins + 0.5 * Number of Ties]/[Total Games Played]). “INDIVPERF” for offensive (defensive) coordinators is the team’s annual percentile rank in points scored (points scored against). Percentile rank, defined as [Raw Rank –1]/[N – 1], normalizes the performance distribution into a 0 to 1 scale with 0 being the worst, 1 being the best, and 0.5 being the mean. Year is equal to the calendar year at the start of the season minus 1970. Predicted probabilities are derived from the logit models by varying INDIVPERF, while maintaining all other variables at their means. Reported are coefficient estimates with estimated standard errors in parentheses. *** (**, *) Significant at the 1% (5%, 10%) confidence level using a two-tailed t-test.
46
Table 9: Management Team Effects
Panel A: Probability of Level-Two Coach Departure Unconditional 31.16% Given No Head Coach Departure 19.48% Given Head Coach Departure 73.68%* Given Head Coach Departure/ Internal Replacement 46.28%* Given Head Coach Departure/ External Replacement 86.21%*† Panel B: Identity of New Level-Two Coaches when Outside Head Coach Hired New Level-Two Coach from Same Previous Team as New Head Coach 19.66% Probability of Hiring From Same Previous Team if Random Hiring 3.51% T-statistic from 2-tailed test of Random Hiring 6.20
Panel A reports the probability of an offensive or defensive coordinator departing his team for any reason in a given year conditional on whether his head coach departs and whether the head coach is replaced by an insider or outsider. Panel B reports the previous team for which new offensive or defensive coordinators worked if new individuals are named to these positions at the time an outside head coach is brought in. The sample of coaches is drawn from all National Football League teams from 1970 to 2001. Team histories are collected from Total Football II: The Official Encyclopedia of the National Football League for the 1970-1998 seasons and from the ESPN.com website (http://msn.espn.go.com/main.html) and the 2000-2002 editions of The Sporting News Pro Football Guide for the 1999-2001 seasons. For each team, we record the identity of the head coach, the offensive coordinator, and the defensive coordinator at the start of each annual season. The primary source of data on coaches’ identities and backgrounds are the teams’ annual media guides archived at the Pro Football Hall of Fame in Canton, Ohio. * Statistically different from No Head Coach Departure sample at the 1% level using a two-tailed t-test. † Statistically different from Head Coach Departure/ Internal Replacement sample at the 1% level using a two-tailed t-test.