Time to Go? Head Coach Quits and Dismissals in Professional Football

That football Head Coaches will be dismissed for poor performance and will quit when they have better outside options seems to be nothing more than a statement of the obvious. But owners may find it hard to distinguish poor performance from bad luck and may find it difficult to identify and attract talented managers from other clubs. Indeed, most of the literature indicates little improvement in team performance when one coach replaces another. Equally, Head Coaches may have few options to move to better clubs even when they are performing well.We identify significant differences between determinants of quits and dismissals that are largely consistent with a standard model which predicts departures occur when the value of the job match specific surplus for one or both parties falls below the value of outside options. However, dismissals and quits are more common in Italy and Spain than in Germany and France, suggesting institutions may be important. We discuss the implications of our findings in the context of principal-agent theory and the wider literature on turnover among CEOs and other corporate leaders.


Introduction
In a principal-agent framework managers are brought in by owners to run their firms with a view to maximising profits. If the firm is a listed company, owners will observe performance annually and, in the light of company performance and ambient labour market conditions, choose how to reward or punish senior management. In the absence of real time performance data and given the costliness of monitoring the activities of senior executives (Bandiera et al., 2012), owners may use proxy measures of corporate performance, such as share price movements, to up-date their information regarding how well senior management are performing. Share price movements and annual profitability may reflect many factors, including changes in market sentiment and changes in business conditions, some beyond the control of the senior executives. Furthermore, even though Chief Executive Officers (CEOs) have an important role in determining the productivity of all other employees due to their position at the apex of the organization (Rosen, 1990), it is extremely difficult to identify the causal impact of leaders on organizational performance. 1 Nevertheless, CEOs are formally responsible for the corporation's performance and may therefore expect that performance to be reflected in their compensation packages and the longevity of any employment contract they may be offered.
When a firm is performing poorly, or more poorly than expected, the CEO can expect to be under pressure to "turn things around" and, if this does not happen, they may be under threat of dismissal (layoff). Poor performance of the firm may be directly related to the decisions or indecision of top executives, or may simply be "bad luck", as in the case of deteriorating market conditions. For this reason, owners often use the firm's performance relative to its competitors to determine executive compensation, thus conditioning on the market conditions all firms in the industry face (Bertrand and Mullainathan, 2001).
Conversely, if a firm is performing very well, other firms may treat this as a signal of the CEO's high ability and seek to poach the CEO. At the very least the CEO may seek to use good performance in on-the-job search to secure a job offer from a better firm. 3 In this paper, we identify the factors that determine the senior executive turnover in a single, global industry where principals receive weekly updates on firm performance. We do so by modelling senior executives' time to exit from a firm, distinguishing between layoffs made by firms and quits, which are employee decisions to leave. The industry consists of small to medium sized businesses offering a single service competing directly against one another in a transparent fashion. Market conditions are very stable over the period of a few years and there are few exogenous factors in this market that can heavily influence firm performance in the short run. So it is, perhaps, unsurprising to find that firm success or failure is often attributed to the CEO.
The industry is professional football and the firms are football clubs. 2 The "CEO" role is performed by Head Coaches -often referred to as "managers" in the UK -who are appointed to run team affairs. Although the scope of the role varies across countries and even within country depending on club owners' preferences, most Head Coaches have the power to recruit football players to the squad, and Head Coaches pick the team for games from that squad. 3 Head Coaches are also responsible for recruiting back room support staff and for the coaching tactics used to beat opponents. It seems reasonable to assume, therefore, that Head Coaches play a crucial role in determining team performance, even though this causal impact has proven rather difficult to identify in practice. 4 Club owners are able to update their information on Head Coach performance with the results from each game, which tend to happen on average once a week during the football season. This provides them with an opportunity to consider Head Coach performance relative to expectations on an almost continual basis, something that is harder to do in circumstances where principals only receive annual financial accounts and find monitoring executive performance costly. 2 In the United States these would be termed "soccer" teams because the term "football" is reserved for American Football. 3 In Continental Europe hiring and release of players is handled by the Director of Football with input from the Head Coach.

4
Whereas football players can only be traded at particular times during the football season, coaches can be laid off or hired throughout the season, as well as in the closed season.
Head Coaches are also able to signal how good they are to prospective employers on a weekly basis through their team's performance, which is often attributed to the Head Coach. Prospective employers are therefore able to update their assessments of Head Coach quality weekly, and may well seek to poach rival teams' Head Coaches, creating strong incentives for Head Coaches who are performing well to quit their existing employer in favour of another, subject to negotiation over early departure clauses in their contracts of employment.
Using a particularly large and rich data set on Head Coaches from the first two tiers of four European leagues over the period 2002 to 2015, we estimate duration models for quits and dismissals. We identify significant differences between determinants of quits and dismissals which are largely consistent with a standard model which predicts that departures occur when the value of the job match specific surplus for one or both parties falls below the value of outside options. However, dismissal and quit probabilities are more common in Italy and Spain than in Germany and France, suggesting institutions may be important.

Theory and Empirical Evidence
In the standard model, workers are hired when the match-specific surplus generated for the firm exceeds the costs of hire. Termination of the contract will occur through dismissal by the employer (often termed "layoff"), or a quit by the worker, where the value of that match for one or both parties falls below the value of an outside option (Farber, 1999). Worker actions such as gross negligence, incompetence or misconduct substantially reduce the net value of the contract to the employer thus resulting in dismissal. Over the life-cycle, a gradual deterioration in worker performance, for example through the degradation of skills 5 or age-related health issues, will reduce the match-specific surplus, especially if the experienced incumbent has benefited from an upward sloping wage profile. 5 Monitoring costs are often too high to establish with any certainty changes over time in the productivity of employees. Exceptions include circumstances in which output is readily identifiable as individual effort, as in the case of academics' publications (eg. Levin and Stephan, 1991). In the case of CEOs, organizational performance is often attributed to them, whether this is justified or not. The costliness of monitoring their inputs means owners prefer to link their compensation to performance outcomes, thus allowing for continual adjustment in the rules governing the sharing of surplus between the principal and agent. Performance pay is akin to wage renegotiation in being able to limit inefficient worker-firm separations (Gielen and van Ours, 2006). Even then the firm must appraise the value of the worker-firm match relative to the value of hiring a new worker.
Firms face the problem that CEOs are heterogeneous in ability and it is hard to identify which are the more talented among them. There is ample evidence that CEOs are heterogeneous in quality and that this affects firm policies (Bertrand and Schoar, 2003).
Furthermore, leaders affect team productivity (Lazear et al., 2015). Muehlheusser et al. (2016) present evidence of substantial heterogeneity in Head Coach ability in the German "Bundesliga", where team performance varies according to the ability of the incoming coach. Theory suggests inefficient hiring in talent markets whereby mediocre workers are re-hired in the face of risk associated with appraising the talent of workers that are new to an industry (Tervio, 2009). This market failure arises where talent is industry-specific, is only revealed on the job and, once revealed, becomes public information. More productive firms hire those revealed to be high-ability whereas less productive firms must experiment with untested new workers. Where there is insufficient discovery of new talent firms are forced to re-hire some workers known to be mediocre. Peeters et al. (2016) confirm that this market failure exists among Head Coaches in professional football in England. 6 We contribute to the literature in three ways. First, our large sample and sizeable number of quits as well as dismissals gives us the power to detect influences on these outcomes that may not have been possible in previous studies. Second, our data contain a richer set of Head Coach characteristics than is commonly available so that we can distinguish various aspects of general and firm-specific human capital, as well as coaches' achievements as football players earlier in their careers. This richer set of covariates offers greater insight into the factors affecting coach exits than has been possible until now. Third, to our knowledge all studies to date focus on single countries. We investigate cross-country differences with data from four countries (Germany, Italy, Spain and France).
The theory and evidence presented above in relation to CEO hires and CEO heterogeneity has implications for Head Coach quits and dismissals in professional football. We use these insights to test five hypotheses with our data. In keeping with this literature we suspect good performance, and performance above expectations, will lower dismissal rates by increasing the net value of the contract to the employer. However, they may also increase the likelihood of a quit due to "poaching" behaviour on the part of competing firms which increases the job offer rate for Head Coaches. Conversely, a sequence of bad results may be perceived as an indicator of poor Head Coach performance, rather than simply a bad run of luck (what Rabin and Vayanos (2010) refer to as a "hot-hand fallacy"), thus raising the likelihood of dismissal and reducing the opportunity to quit for another job. Conditional on experience and performance, we anticipate that older Head Coaches are more likely to be dismissed due to employer expectations of a reduction in future matchspecific surplus arising from risks such as health issues, and the relative costliness of Head Coaches late in their career. We anticipate that general human capital such as coach experience will be valued by employers, potentially protecting the Head Coach from dismissal, even when performance is below expectations (although this is not what Van Ours and Van Tuijl (2016) find in their study). Similarly, signs of success in the coach's previous jobs (such as winning cups or titles, or getting clubs promoted) will delay the point at which the employer dismisses a coach conditional on performance.
8 It is less clear that experience and past performance will affect quit probabilities since job offer rates are likely to be driven by performance in the current job although, given the likely market value of past coaching experience and past coach performance in tackling management problems in a new environment, one would expect general human capital to raise job offer arrival rates, potentially increasing quits. However, older Head Coaches are less likely to attract outside offers since prospective employers will anticipate a lower job match surplus from older employees, all else equal.
Hypothesis Three: Increased firm specific human capital will lower dismissal probabilities but will have no significant effect on quit probabilities Increases in firm specific human capital should lower dismissal probabilities by providing workers with insights about the specifics of the firm and its production processes which can then raise labour productivity (Becker, 1962). Of course, human capital investments such as on-the-job training are potentially endogenous with respect to turnover probabilities since employers will not make these investments without appraising the likelihood of a return (Becker, 1962: 19-20). To overcome this problem we focus on measures of firm-specific human capital pre-dating the current spell, namely number of previous spells at the club, whether the Head Coach was hired from within, and whether the Head Coach was an explayer at the club.
The effects of these types of firm specific human capital on quit probabilities are less clearcut, a priori. On the one hand, even if alternative employers partly discount the value of investments made in another firm, they may nevertheless view these mutual employerworker investments as a signal of the worker's worth in the labour market, in a similar way to internal promotions (De Varo and Waldman, 2012), potentially raising the probability that a Head Coach will receive an outside offer. On the other hand, a Head Coach's firmspecific human capital investments are likely to lower quit rates to the extent that they increase the value of the current job match relative to outside options, thus raising the probability that the Head Coach will reject any outside offers (Stevens, 2003).

Hypothesis Four: Conditional on performance the Head Coach's experience as a
professional footballer will not significantly affect dismissal or quit probabilities Goodall et al. (2011) argue that brilliance as a basketball player is a good predictor of subsequent performance as a team coach. Past playing experience facilitates a teaching role for Head Coaches. Moreover, a Coach's motivational effort may well be more credible in the eyes of current players if the coach has had success playing the sport. This may explain why leaders with expert knowledge of an organization's core business improve its performance (Goodall and Pogrebna, 2015). However, having conditioned on team performance under the Coach, it seems unlikely that the Head Coach's experience as a footballer will influence quit and dismissal probabilities, unless it increases the playing squad's respect for the incumbent manager, regardless of performance, thus "keeping the dressing room" for longer than might otherwise be the case.

Hypothesis Five: Dismissal probabilities will be greater in Italy and Spain than they are in Germany and France
Cross-country differences may emerge for a number of reasons. First, labour markets operate quite differently due to differences in employment protection, for example (European Commission, 2013). Second, football institutions differ across countries in terms of governance structures, the financing of clubs, media attention, and team ownership. Third, there are cultural differences in attributes such as patience that might mean principals' responses to agents' performance differs across countries (Dohmen et al., 2015). Although we do not observe these differences directly we can examine crosscountry differences with country dummy variables. Our expectation is that turnover rates are likely greater in Italy and Spain than in Germany and France, in part because pressure on owners to act in the face of poor performance from the media and fans is arguably greater in the former two countries. Governance structures in Germany, and to some extent France, where clubs tend to be member-owned and less open to foreign investment than those in Italy and Spain, may predispose owners to take a longer-term view (Hall and Soskice, 2001). Furthermore, German and French individuals exhibit greater patience than Spaniards and Italians in social surveys, suggesting that clubs with member-owners may be less likely to dismiss Head Coaches conditional on performance (Falk, 2016).
It is less clear whether there will be cross-national differences in quit probabilities but these may occur if Head Coaches themselves exhibit different levels of patience, or some national labour market conditions or institutions are more conducive to labour mobility than others.

Data and Empirical Approach
We we observe the start date for all coaches' initial employment spells, including those that overlap the start of the initial football season in our data.
First, we estimate time to exit (whether a dismissal or quit), time to dismissals and time to quits separately using Cox proportional hazard rate models estimated with maximum likelihood. The 693 Head Coaches in our data experience a total of 1,758 exits. These are multiple-failure per subject data: the minimum number of exits for coaches is zero, the maximum is 11, the median is 2 and the mean is 2.5. Time is measured in days. The Having dropped cases with missing observations our exits, estimates are run on 638 Head Coaches. The models are run with a robust estimator to account for heteroscedasticity and standard errors are clustered to account for the non-independence of Head Coach spells.
We run an identical model specification across all three dependent variables. The Following Stadtmann (2006)  international experience as a coach, a dummy variable identifying whether he is currently working in his native country, whether he has coached a team that was promoted to a higher division in the past, whether he has coached a team that was relegated to a lower division in the past, whether he has coached a team that has won a cup before, and whether he has coached a team that has won the league title. The second set contains three measures of firm-specific human capital, namely whether the Head Coach ever played professional football for the club he is now coaching, whether he has previously coached the club he is currently coaching, and whether he was an internal appointment from within the club.
The third set of variables containing coach characteristics captures his ability as a professional football player, in the spirit of Goodall et al. (2011) who, as noted earlier, emphasise the value of past experience playing a sport for success as a coach in that sport.
These variables are whether the coach ever played football for his national side, number of years' experience as a professional football player, a dummy variable identifying whether the coach had ever played in the top division as a professional footballer and a dummy identifying Head Coaches who had never played football professionally.
Models also capture real time as indicated by the football season, dummy variables capturing the country league, and a dummy identifying whether the games are taking place in the top or second tier of football in that country.
We also run a discrete time competing risks model since a non-censored spell can end with one of two occurrences, namely a dismissal or a quit. Finally we estimate mixed effects parametric survival models which account for unobserved heterogeneity across Head Coaches. 8 We run separate models for leaving, dismissals and quits. The conditional distribution of the outcomes given the random effects is set to a Weibull distribution. Descriptive data for the variables used in the analysis are presented in Appendix Table 1. In keeping with Hypothesis One improved performance and greater performance above expectations reduce the likelihood of Head Coach dismissal. However, contrary to Hypothesis One, neither performance variable was significantly associated with the likelihood of quitting. One possible reason for this, discussed earlier, is that good performance triggers performance-related pay in Head Coaches' contracts reducing the value of accepting outside offers. Alternatively, Head Coach contracts may be "stickier" than one might think, with teams waiting until the end of the season before they make changes, perhaps because this is when many contracts cease in any event and employers are keen to limit financial and other liabilities which might result in "poaching" another team's manager, even if this appears optimal. Support for this alternative "stickiness" hypothesis is indicated by the positive and statistically significant coefficient on the dummy variable identifying the last day of the season in both the quits and dismissals models, suggesting 7 This imposes a Weibull distribution on the data which seemed reasonable having experimented with alternatives since log(time) is on the margins of statistical significance in estimating time to dismissal. 8 We use the STATA routine MESTREG.

Results
14 that both teams and Head Coaches often simply wait until the season ends before making changes. 9 Running counter to this is the negative and statistically significant coefficient on the games left to the end of the season, which suggests the likelihood of both quits and dismissals rises as the season end approaches, probably because teams seek to make changes at the "business end" of the season when the consequences of failure or success become increasingly apparent. 10 The models offer strong support for the proposition in Hypothesis Two that general human capital is valued by employers and protects them from dismissal even conditioning on performance and performance expectations. The nine general human capital covariates are jointly statistically significant in the dismissals model (chi-sq(9)=47.05, p>chi2=0.0000).
Indicators of previous success (coaching a team that was promoted to a higher division or won a cup) and Coach experience reduce the likelihood of dismissal. So too does working in one's home country, perhaps because "local" knowledge of club competition provides Coaches with insights into how to manage teams, or else because it improves communication between the Coach and players, most of whom are usually drawn from the home country. 11 On the other hand, as anticipated in Hypothesis Two, Head Coach age is positively correlated with coach dismissal, suggesting employers expect future job-match surplus to fall as workers get older.
The general human capital variables are also jointly statistically significant in the quit models (chi-sq(9)=20.82, p>chi2=0.0135). As anticipated, greater coaching experience is associated with a greater likelihood of quitting, as might be the case due to "poaching".
However, having more previous spells as a Head Coach reduces the likelihood of quitting, as one might expect in a "job shopping" model in which workers search until they achieve a 9 Contracts often expire at a season's end, so some of these departures will reflect the non-renewal of fixed term contracts. 10 Owners may be less concerned by early poor performance if they think there is sufficient time left in the season for a coach to "turn things round". Lower quit rates earlier on may also reflect the relative paucity of available job slots, thus limiting the job offer arrival rate. 11 Coaches often create a bond with the team's supporters who can express their support for the Coach vocally at games. The degree of support for a Coach may vary between native-born and foreign Coaches, and may sway principals when deciding whether to dismiss a Coach. The bond may also increase the value of remaining at the club, lowering quit rates. good job match. Previous experience coaching a team that was subsequently relegated is also associated with lower quit probabilities, consistent with Hypothesis Two's assumption that prospective employers will be less likely to make job offers to coaches who have previously been associated with failure.
In Hypothesis Three, we proposed that greater firm specific human capital should lower dismissal probabilities but would have no significant effect on quit probabilities. This proves to be the case. The three firm-specific human capital variables are jointly statistically significant for dismissal (chi-squared(3)=10.18, p>chi2 0.0171), although the only measure that is statistically significant on its own is the number of previous spells coaching the club, which is strongly associated with a lower likelihood of being dismissed.
In Hypothesis Four we suggested that, conditional on performance, the Head Coach's experience as a professional footballer would not significantly affect dismissal or quit probabilities. The four measures capturing Head Coaches' experience as professional football players were not significant in the quit models and they were jointly nonsignificant (chi-squared(4)=1.60, p>chi2=0.8096), but they were jointly significant in the dismissal models (chi-squared(4)=8.77, p>chi2=0.067), driven by the positive effect of never being a professional footballer on dismissal probabilities. It is possible that intimate knowledge of the game, acquired when playing as a professional, gives Head Coaches tacit skills which are valued by the principal, such as the ability to communicate with players or motivate them. Alternatively, this effect may capture the difficulties Head Coaches have in commanding respect in the dressing room if they have not played the game.
In Hypothesis Five we suggested there would be differences in Head Coach turnover across the four countries in our data. The survival curves for leaving, dismissals and quits for the raw data confirm that there are cross-country differences (Figures 1, 2 and 3). Figure 1 shows the survival rate is lowest in Italy, followed by Spain, Germany and finally France.
When looking at dismissals only ( Figure 2) the country ordering is the same but the dispersion between countries is larger. The ordering for quits is a little different, with survival rates longest for Head Coaches in Germany (Figure 3).
[INSERT FIGURES 1, 2 AND 3] When we adjust for other factors in Table 1 the model confirms that both dismissal and quit probabilities are highest in Italy. Head Coaches in Spain also have higher dismissal probabilities than those in France, and Head Coaches in Germany are the least likely to quit. The fact that we see cross-country differences in quit as well as dismissal rates suggests that the cross-country differences are unlikely to simply reflect country differences in the pressures facing football club owners as principals, or differences in governance arrangements although, of course, these may be very important in explaining cross-country differences in dismissal rates. There is no differential in the quit or dismissal probabilities for Head Coaches in tiers 1 and 2.
[INSERT TABLE 2] Table 2 presents similar analyses but this time models quit and dismissal probabilities simultaneously in a discrete time competing risks model. The coefficients are presented as relative risk ratios compared to the no exit baseline. Before commenting on the hypotheses outlined above it is notable that there is duration dependence in dismissals but not quits.
That duration dependence follows a Weibull distribution, rising a little initially then falling over time. In the CEO literature duration dependence has been equated with CEOs' abilities to "capture" corporate governance structures, thus limiting stakeholders' ability to remove senior executives (Gregory-Smith et al., 2009). This is unlikely in our setting. Instead it may reflect unobserved heterogeneity across Head Coaches, such that only the best coaches are observed in long duration jobs, or else longevity reflects the underlying quality of the original job-match (true duration dependence). We return to this below.
[ INSERT TABLE 2] The competing risks model also clearly quantifies the size of some of the time-varying covariates relative to the baseline of no departure. For instance, it is apparent that dismissals are more sensitive to the number of games left in the season than quits, while the relative risk of quits is higher than that for dismissals in the closed season.
Turning to our hypotheses, with just a few exceptions the discrete time competing risks model confirms the findings from the separate Cox proportional hazards quit and dismissal models. However, there are some important differences that are noteworthy. First, contrary to Hypothesis One, the discrete time competing risks model indicates that good performance and performing beyond expectations were strongly and significantly associated with lower quit probabilities, rather than higher quit probabilities. The implication is that Head Coaches are able to use their strengthened bargaining hand to renegotiate their compensation package with their existing employer, rather than quitting for a higher wage, or else performance pay clauses are triggered in their contracts increasing the value of the existing contract relative to potential outside offers which, as noted earlier, is akin to a wage renegotiation limiting inefficient worker-firm separations.
The other noteworthy difference is that the cross-country differences in both dismissals and quits are a little more marked in the competing risks models than in the Cox proportional hazard models. This, together with the fact that both quit and dismissal probabilities are higher in tier two than they are in tier one in the competing risks models, underscore the importance of differences in departure probabilities across leagues.

[INSERT TABLE 3]
Finally in Table 3 we present mixed effects parametric survival models for leaving, dismissals and quits separately. We assume an underlying Weibull distribution to departures. These are random effects models fitting coach-specific intercepts to account for unobservable differences across Head Coaches. Although some individual coefficients lose their statistical significance the estimates generally support the results presented in the Cox proportional hazards and competing risks models.
In relation to Hypothesis One, there is further evidence that good performance and performance beyond expectations are both significantly associated with lower dismissal probabilities. As in the competing risks model there is also statistical support to indicate that good performance is associated with a lower probability of a Head Coach quitting, but this is not true for performing beyond expectations.
Hypothesis Two also receives strong support. The variables capturing general human capital are jointly statistically significant in both the dismissal (chi2(9) 27.75 p=0.0011) and quit models (chi2(9) 17.19 p=0.0458). Individual coefficients have the anticipated signs. Dismissal probabilities rise with age, as predicted, but quits do not.
There is also broad support for Hypothesis Three. Firm specific human capital variables are jointly statistically significant for dismissals (chi2(3) 10.99 p=0.0168) with the one strong significant effect being the negative effect of previous spells of coaching at a club on dismissal probabilities. In contrast firm-specific human capital variables were only jointly on the borderline of significance for quits (chi2(3) 7.71 p=0.0524) and none of the variables were individually significant.
As predicted in Hypothesis Four, experience as a professional player was not generally significant for dismissal (chi2(4) 8.10 p=0.0881) and quit probabilities (chi2(4) 1.17 p=0.8833) having conditioned on performance. The exception was the positive effect that never having played professional football had on dismissal probabilities. Finally Hypothesis Five is supported again, with Head Coaches in Spain and Italy being much more likely to be dismissed or quit relative to those in France and Germany.

Conclusion
Understanding the value of job-matches and the factors leading to their cessation is fundamental to the nature of labour markets. Although initial efforts distinguishing between the determinants of quits and dismissals go all the way back to Farber's (1980) seminal paper, there has been little research on the relative risks of quits and dismissals in the intervening period. Instead, analysts have focused their attention on various routes in and out of unemployment, no doubt driven by social welfare concerns to minimise exposure to unemployment. 12 The exceptions relate to research on CEOs in public listed companies and the fortunes of sports Head Coaches. Even here it has proved difficult to make the distinction empirically because it is not usually obvious whether a worker has been dismissed or left voluntarily and most data sets are too small to identify with confidence the factors determining the small number of departures which are quits.
Nevertheless, it is clear from this small body of empirical research that quits and dismissals are very distinct forms of separation. It is also well-established that poor performance substantially raises the probability of dismissal. Expectations regarding performance also play a role. For instance, performing below expectations signalled by the betting market predicts dismissal. There are also a number of contradictory findings in the literature, and uncertainty surrounding some issues such as the role played by general and firm-specific human capital.
Using a particularly large and rich data set on Head Coaches from the first two tiers of four European leagues over the period 2001 to 2015 we estimate Cox proportional hazards and discrete time competing risks models for quits and dismissals. We can contribute to the literature because ours is a particularly large and rich data set, containing many variables that do not normally appear in the literature. The many quits we have in the data mean we have the statistical power to identify effects of covariates that may not have been apparent in previous studies. We identify significant differences between determinants of quits and dismissals that are largely consistent with a standard model which predicts departures occur when the value of the job match specific surplus for one or both parties falls below the value of outside options. However, dismissals and quits are more common in Italy and Spain than in Germany and France suggesting that institutions and corporate governance may be important.      (1) Significance Key: * p<0.10, ** p<0.05, *** p<0.01 (2) t-statistics in parentheses (3) 68,208 manager-game observations.