High Performers = Better Leaders? Evidence From 55 Years of Professional Soccer on the Validity of Performance-based Promotion to Leader Positions

Following the logic of the performance requirements perspective (see Zaccaro et al., 2018), which relies on individual differences to predict LP, performance-based promotion to leader positions should be a valid strategy only to the extent that the employee and the leader position have matching performance requirements (Asher & Sciarrino, 1974; Zaccaro et al., 2018). General taxonomies for the workforce (e.g., Bartram, 2005) as compared to managerial taxonomies (e.g., Borman & Brush, 1993; Tett et al., 2000), however indicate a low match between employee and leader tasks (Schleu & Hüffmeier, 2021): Many dimensions of the managerial taxonomies are rather specific to leader positions (such as guiding, directing, coordinating, and motivating subordinates; Borman & Brush, 1993; Tett et al., 2000), indicating a low match of performance requirements. Moreover, leader positions as compared to employee positions are more complex (Hunter et al., 1990) and require “a variety of different activities in carrying out the work, which involve the use of a number of different skills and talents” (Hackman & Oldham, 1975, p. 161).

Further, to compare the performance requirements of employee and leader positions in more detail, it is instructive to focus on commonly accepted predictors of leadership success (for an overview, see Zaccaro et al., 2018), such as personality, job knowledge, and motivation. Concerning personality predictors, the link to leadership success seems to be context-dependent (see De Hoogh et al., 2005) and contingent on employed performance criteria (e.g., rated leader performance versus team performance; for meta-analytic evidence, see DeRue et al., 2011). Taking into account the variability of required personality traits (i.e., for different leader positions), the performance requirements of employee and leader positions concerning personality may match only in some specific contexts. That is, while certain attributes influence EP positively (e.g., conscientiousness or agreeableness) they might not necessarily translate into high LP—but could in some cases even translate into low LP. For instance, a successful, dutiful employee might become an ineffective micromanager or a conflict-avoiding supervisor (see Smith et al., 2018).

Job knowledge acquired as an employee might be helpful when becoming a leader, for instance, for structuring the employees’ work (see Day et al., 2009). At the same time, previous experience might not translate to new positions or contexts (see Salomon & Perkins, 1989; van Iddekinge et al., 2019) and expertise might reduce a leader’s cognitive flexibility facing changes (see Dane, 2010). Furthermore, the need for specific job knowledge acquired in an employee position should decrease with higher hierarchy levels (see Day et al, 2009; Katz & Kahn, 1978). Thus, the performance requirements concerning job knowledge again may match for some contexts (e.g., lower-level leadership positions), but not in general.

While the importance of motivation for performance is obvious and was recently reaffirmed (for meta-analytic evidence, see van Iddekinge et al., 2018), the manifestation of employee motivation could either be stable (i.e., more trait-like) or context-dependent (i.e., more state-like). Consequently, it is unclear to what extent motivation in an employee position transfers to leader positions (i.e., motivation to lead; Kark & van Dijk, 2007) due to the different tasks and responsibilities (Porter et al., 2016). To sum up, the performance requirements perspective allows predictions of positive (i.e., for highly overlapping performance requirements), null (i.e., for little overlap of performance requirements), and negative relations (i.e., if performance requirements of the employee position hinder high performance as a leader) between EP and LP when considering the specific positions. On average, the explanatory value of EP for LP should be low, as the performance requirements of employee and leader positions match only to a limited degree.

We further consider the follower-centric perspective to understand potential links between EP and LP (see Steffens et al., 2021; Uhl-Bien et al., 2014). According to this perspective, leadership is a social construction of followers and it focuses on the requirements and processes to convince a team to follow (Lord & Maher, 2002; Lord et al., 1984) or to gain credibility in a team (Kouzes & Posner, 2011). Followers evaluate their leaders (see Lord & Dinh, 2014) based on past experiences and their socialization concerning typical characteristics of leaders (i.e., implicit leadership theories [ILTs]; Phillips & Lord, 1981; Schyns et al., 2005). Based on their experiences, followers make sense of organizational processes (Weick, 1995): Prior high performance as an employee should increase the degree to which leaders are perceived to fulfill ILTs (Ensari & Murphy, 2003; Lord & Maher, 2002). This might facilitate the attribution of positive outcomes to them (e.g., Meindl, 1995).

Leaders who were promoted based on previous EP in particular should be perceived to embody core attributes of the team (i.e., be prototypical; Schleu & Hüffmeier, 2021). As a result, the leader has informative value for the team and might be perceived as a role model (Hogg, 2001; Uhl-Bien et al., 2014). Consequently, followers attribute higher credibility to prototypical leaders and support them more (Uhl-Bien et al., 2014), which should result in more successful leaders (i.e., higher LP; Steffens et al., 2021). Thus, the follower-centric perspective suggests a moderate, positive link between EP and LP.

The propositions of TEL (Goodall & Bäker, 2015) emphasize the importance of expert knowledge (i.e., acquired as an employee) to be a good leader, at least for knowledge-intensive organizations. Expert knowledge is assumed to be acquired through technical education, practice, and working experience in a particular sector. Goodall and Bäker (2015) assume that expert knowledge is beneficial for LP due to the following main reasons: First, it is assumed to provide a particularly solid base for decision making (see Goodall & Bäker, 2015). Thus, expert leaders as compared to non-expert leaders profit from representing information holistically and rely on abstract concepts to solve problems (for a related overview, see Ericsson et al., 2006). Second, expert leaders should have a good understanding of the core workers’ values, motivations, and typical challenges, due to a shared knowledge base with these workers. As they have a common background with their subordinates, they are potentially perceived “as being the first among equals” (Goodall & Bäker, 2015, p. 57, note that this aspect is somewhat comparable to being prototypical). This constellation should enable them to create a productivity-enhancing working environment, to set realistic goals, and to evaluate the performance of employees appropriately. Third, Goodall and Bäker (2015) emphasize the advantages of expert leaders in personnel selection due to the homophily principle (cf. Rogers & Bhowmik, 1970): When expert leaders hire new personnel, they are assumed to hire applicants with a high potential (similar to themselves), instead of feeling threatened by capable applicants. This should improve the quality of personnel selection compared to non-expert leaders. In sum, TEL would assume that newly promoted leaders profit from their expert knowledge and, hence, that there is at least a moderate, positive link between EP and LP.

The three theoretical perspectives outlined above make contrasting predictions concerning the predictive validity of EP for LP initially and over time. According to the performance requirements perspective, EP should on average have limited explanatory value for predicting LP, as the performance requirements of employee and leader positions differ. Consequently, the predictive validity of EP for initial LP (directly following the promotion [at t1]) should be small at most (Hypothesis 1a).¹ As we report standardized path coefficients, we used conventions for correlation coefficients as a yardstick (i.e., small: 0.10; moderate: 0.30; large: 0.50; Acock, 2014; Cohen, 1988); thus, the proposed relation in H1a corresponds to a standardized path coefficient ranging from -0.10 to 0.10. Further, we consider the effect of time on the relationship between EP and LP: Based on the performance requirements perspective, LP may change over time due to learning and adaptation to the requirements of the leader position. However, these processes are not necessarily related to previous EP. According to this perspective, the predictive validity of EP for LP should remain small over time (Hypothesis 1b).

Based on the follower-centric perspective, prior high EP should increase the degree, to which leaders are perceived as prototypical and to which they meet ILTs (Ensari & Murphy, 2003; Lord & Maher, 2002). Hence, followers should support promoted leaders more (Uhl-Bien et al., 2014), which should result in more leadership success (i.e., higher LP; Steffens et al., 2021). Thus, the predictive validity of EP for initial LP (directly following the promotion [at t1]) should at least be moderate (Hypothesis 2a), corresponding to a standardized path coefficient ≥ 0.30. Over time, leaders do not only need to embody core attributes of the team to be perceived as prototypical, but also need to prove their prototypicality through their engagement for the team (for instance by empowering team members; for pertinent overviews, see Haslam et al., 2011; Steffens et al., 2014). However, neither the follower-centric perspective nor empirical evidence (that we are aware of) suggest a link between former EP and leaders’ engagement for their team. Hence, while the follower-centric perspective proposes a link between EP and leaders’ initial perceived prototypicality (and thus initial LP), the effect of EP should diminish over time: Leader engagement for the team becomes more important for leaders to maintain their perceived prototypicality over time, which consequently affects the experienced support by their team, and, thus, their success as a leader (Haslam et al., 2011; Steffens et al., 2014). In addition, neither higher prototypicality nor identity leadership propose a link to the quality of decision making (e.g., developing strategies); thus, over time, the team may also be confronted with bad decisions. Consequently, the predictive validity of EP for LP should decrease over time (Hypothesis 2b).

Following the logic of TEL, leaders should benefit from their expert knowledge when getting promoted (e.g., by making better decisions and creating a productivity-enhancing working environment). Thus, the predictive validity of expert knowledge—as indicated through EP—for initial LP (directly following the promotion [at t1]) should at least be moderate (Hypothesis 3a), corresponding to a standardized path coefficient ≥ 0.30. Despite one similar mechanism between the follower-centric perspective and TEL (i.e., having a common background with subordinates is somewhat comparable to being prototypical), both theoretical perspectives suggest different developments of the EP-LP link over time. While we argued above that the follower-centric perspective assumes that mechanisms unrelated to EP will become more relevant over time (i.e., engagement for the team), TEL incorporates mechanisms, which are supposedly related to EP and should unfold over time. That is, expertise acquired as an employee will help leaders, for instance, to hire applicants with high potential and create a productivity-enhancing working environment (Goodall & Bäker, 2015). With time, the overall effect of the proposed mechanisms thus should increase. According to TEL, the predictive validity of EP for LP should increase over time (Hypothesis 3b).

As explained above, the performance requirements of employee and leader positions generally do not match well (cf. Bartram, 2005; Borman & Brush, 1993; Tett et al., 2000). Among other aspects, leader positions are more complex than employee positions (Hunter et al., 1990). Following the logic of the performance requirements perspective, more complex employee positions may approximate the complexity of later leader positions (Hunter et al., 1990). Consequently, (high) EP in complex positions (i.e., high variability in tasks) should indicate the ability to handle various activities and a diverse skill set, which should then predict (high) LP, as this is a relevant leader characteristic. Therefore, the predictive validity of EP for LP should be higher for more complex employee positions (see Quińones et al., 1995; Tesluk & Jacobs, 1998; Hypothesis 4), corresponding to a standardized path coefficient > 0.10. In comparison, the other two outlined theoretical perspectives are rather mute about the proposed moderation (i.e., complexity of the employee position).

To provide another test of the outlined theoretical perspectives, we also derive contrasting predictions on having expertise as an employee versus not having such pertinent expertise. Specifically, we also test whether our previous predictions (i.e., H1a-b, H2a-b, and H3a-b) will hold if experience at a pertinent employee position (yes vs. no) instead of EP is used as a predictor of LP. As the overlap of performance requirements of employee and leader positions is on average rather limited (see Bartram, 2005; Borman & Brush, 1993), the performance requirements perspective predicts the following: Leaders, who have pertinent experience as an employee, should not show better LP than leaders without such experience and this should not change over time (Hypothesis 1c).

According to the follower-centric perspective, a leader’s previous experience as an employee increases perceived prototypicality and follower performance (Steffens et al., 2021; Uhl-Bien et al., 2014). Experience as an employee, however, should not relate to other aspects of perceived prototypicality (e.g., empowering team members; see Haslam et al., 2011), thus its effect should decrease over time: Leaders, who have pertinent experience as an employee, should initially show better LP than leaders without such experience. This difference should, however, decrease over time (Hypothesis 2c).

TEL (Goodall & Bäker, 2015), in contrast, predicts that leaders with employee experience should have a better start (e.g., by creating a productivity-enhancing working environment) and this difference should become even more pronounced over time (e.g., due to the lagged consequences of better personnel selection decisions). Leaders, who have pertinent experience as an employee should, thus, initially show better LP than leaders who have no such experience. This difference should further increase over time (Hypothesis 3c).

We included all head coaches (i.e., the population) of clubs competing in Germany’s first male soccer league (i.e., Bundesliga). Thus, we went through every season of the Bundesliga from 1963/64 (i.e., its first season) to 2018/19 on the homepage www.​transfermarkt.​de, identified all head coaches (N = 407) of the competing clubs, and extracted their data. We attempted to obtain comprehensive data on their career paths. To do so, we included player and coach performance data from the first and second German soccer league, as well as data from every first and second foreign league included in the Union of European Football Associations (UEFA), if the data were available on the respective websites (for further details, please see below).³ About ¾ of the coaches were former professional soccer players (n = 309; 75.92%). Among those, 222 played in the Bundesliga or a first soccer league of the UEFA, 17 played in Germany’s second soccer league, and 70 coaches played only in lower soccer leagues or detailed information about their professional player career could not be obtained. The remaining 98 coaches (24.08%) had no professional player experience. Coaches with professional player experience participated on average in 143.18 matches of the Bundesliga (SD = 140.97) during their player careers.

As we gathered data on the whole population of head coaches of the Bundesliga between 1963/64 to 2018/19, we examined in a first step the mere effect size of the EP-LP relationship without statistical inference. Note that we nevertheless go beyond mere effect sizes in subsequent steps (Alexander, 2015) since we intend to make inferences on the EP-LP relationship outside the Bundesliga context and to assess the EP-LP link while accounting for potential control variables. In a second step, we therefore gathered data on potential control variables (i.e., players’ injuries, age, the average time of the player rating, team quality [before t1, t2, and t3], and continuous employment [t1-t2 and t2-t3]) and tested their relationship with our indicators of LP (i.e., the dependent variable). Based on the recommendations of Becker (2005), we only added variables as controls to our main analyses if they showed a significant relationship with the dependent variable (i.e., we added team quality [before t1, t2, and t3], continuous employment [t1-t2 and t2-t3] for our main analyses; for bivariate correlations, see Table 2). Next, we analyzed our panel data with manifest path models (including relevant control variables) and conducted six separate analyses for the different operationalizations of player and coach performance.⁷ In particular, we included the hypothesized paths from EP to LP (at t1, t2, and t3) and added complexity and the respective interaction term (i.e., EP × complexity) to our model (see Fig. 1). Furthermore, we included paths from LP at t1 to t2 and from t2 to t3. We also assumed that LP will be further predicted by our control variables (i.e., team quality [before t1, t2, and t3], and continuous employment [t1-t2 and t2-t3]) and specified our model accordingly. We also allowed for covariances (see Fig. 1, for illustrations of our model specifications in model 1). In a third step, we conducted additional analyses to check for the robustness of our results (please see below).

When analyzing the link between EP and LP (Hypotheses 1–4), we only considered coaches who had been soccer players before (n = 309). Similarly, we included only players in our analyses who were graded (n = 181) for the respective analyses (i.e., to ensure there is no missing data concerning the predictor variable; in this case, missing data for LP or control variables (e.g., prior team quality) were estimated using the FIML estimator; Enders, 2010). To test for changes in predictive validity of player performance for coach performance over time, we relied on model comparisons: We computed a restricted model where we restricted the path coefficients of the player performance to be equal over time and compared this model to the original model (i.e., with free path coefficients). Then we conducted χ² difference tests to investigate whether the restriction impaired the model fit. Furthermore, we also conducted analyses considering the aggregated career of the coach (i.e., by considering the aggregated coach performance) to test for the overall effect of performance-based promotion.

To account for different metrics, we standardized the predictor and criterion variables. More specifically, we standardized the coach performance from all measurement points simultaneously to allow for comparisons over time. As effect size measure, we report standardized path coefficients and used the following yardstick for their interpretation: small effects correspond to path coefficients of 0.10, moderate effects to coefficients of 0.30, and large effects to coefficients of 0.50 (Acock, 2014; Cohen, 1988). To correct for alpha inflation associated with multiple tests, we adjusted the p-values for each model through Benjamini and Hochberg’s (1995) step-up Bonferroni procedure as recommended for structural equation models (Cribbie, 2007). In the following, we report p-values that are adjusted in accordance with this procedure.

In a first step, we examined the link of the EP-LP relationship in the population of Bundesliga coaches without statistical inference (and without considering any control variables). Bivariate correlations indicated at most a small to moderate link between EP and LP, both initially (t1) and over time (t2 and t3). At t1, the correlations between EP and LP ranged from rs ≥ .01 to rs ≤ .15, with an average correlation of r = .06 and only one of six coefficients exceeding the threshold for small correlations. At t2 and t3, the correlations between EP and LP were in a similar range (t2: rs ≥ -.03 to rs ≤ .18, with an average correlation of r = .09; t3: rs ≥ -.06 to rs ≤ .07, with an average correlation of r = -.01). Across t2 and t3, only three out of 12 coefficients exceeded the threshold for small correlations (for all correlations, see Table 2). Both the complexity of the player position (rs ≥ -.06 to rs ≤ .07, with an average correlation of r = .00), as well as the interaction-terms of complexity and EP indicated at most a small to moderate relation (rs ≥ -.12 to rs ≤ .17, with an average correlation of r = .04) with LP initially and over time (see Table 2). Finally, we also looked at the relationships between player experience and LP, initially (rs ≥ -.04 to rs ≤ .07, with an average correlation of r = .02) and over time (t2: rs ≥ -.04 to rs ≤ -.01, with an average correlation of r = -.03; t3: rs ≥ -.04 to rs ≤ -.01, with an average correlation of r = -.03).

Overall, these correlations are mostly in line with the performance requirements perspective (Hypothesis 1a-c). As the results of the different analyses include some variance (see Table 2), we will rely on the following, more complex analyses (see the next paragraph) to draw firmer conclusions. In particular, we will include control variables (e.g., prior team quality of the coached team) to provide more robust tests.

We tested six different path models (for model fit statistics, see Table 3) to realize all combinations of different player performance measures (i.e., number of played matches vs. average player rating) with coach performance measures (i.e., PPG, goal ratio, and the number of coached matches).⁸ As our analyses revealed a consistent pattern of results (for more information see Table 3), we exemplarily describe the findings of the first model in more detail: The first model (see Fig. 1) indicated no significant link between the number of played matches and coach performance (i.e., operationalized as PPG) directly after the promotion to the coach position (t1: β = -0.00, SE = 0.06, p = .986). Over time, the relationship remained non-significant (t2: β = 0.02, SE = 0.06, p = .965; t3: β = -0.04, SE = 0.06, p = .810). To test for changes in the predictive validity, we restricted the path coefficients of player performance to be equal over time and conducted a χ² difference test between restricted and unrestricted models (see Table S1 in the supplemental file for all restricted models). The results indicated no difference (Δχ² = 0.48, Δdf = 2, p = .788). The path models with other combination of EP and LP indicators (see Table 3 and Table S1, Models 2–6) mirror those findings. Overall, the results pattern, thus, supports Hypothesis 1a and 1b, as the relationship between player performance and coach performance was at most small and non-significant initially (at t1) as well as over time (at t2 and t3). Further, our results did not indicate a significant change of the relationship between player and coach performance over time.

Further, we also tested the relation between player performance and overall coach performance (i.e., aggregated over the whole coach career) with six models without the FIML estimator (for model fit statistics, see Table 4). Model 1-6 indicated no significant link between player performance (operationalized via the number of played matches or the average player rating) and coach performance (operationalized via PPG, goal ratio, or the number of coached matches). Overall, those aggregated analyses indicate a small-to-moderate, non-significant relation between player and coach performance.

We inspected the hypothesized interaction of player performance and job complexity in our prior analyses (see Table 3 and 4). The main effects of job complexity (except for one analysis; see Table 4) and the interaction effects (i.e., player performance × job complexity) were insignificant, both initially and over time. Thus, our analyses did not support Hypothesis 4.

Finally, we tested the link between player experience (professional soccer player: yes vs. no) and all three operationalizations of coach performance over time with three models (for model fit statistics, see Table 5). The first model did not indicate a significant link between player experience and coach performance (i.e., operationalized via PPG) initially after the promotion to the coach position and over time (t1: β = 0.07, SE = 0.11, p = .137; t2: β = 0.02, SE = 0.12, p = .756; t3: β = -0.03, SE = 0.11, p = .756). To test for changes in the predictive validity, we restricted the path coefficients of player experience to be equal over time and conducted χ² difference tests. The results indicated no difference (Δχ² = 1.98, Δdf = 2, p = .371). The second model indicated a small significant relationship between player experience and coach performance (i.e., operationalized via goal ratio) directly following the promotion to the coach position, t1: β = 0.09, SE = 0.12, p = .038. Over time, the relationship became non-significant (t2: β = 0.03, SE = 0.09, p = .591; t3: β = -0.05, SE = 0.12, p = .433). Testing for changes in the predictive validity, the results indicated no change (Δχ² = 4.27, Δdf = 2, p = .118). The third model neither indicated a significant relationship between player experience and the number of coached matches directly after the promotion to the coach position (t1: β = -0.03, SE = 0.10, p = .561) nor later (t2: β = 0.08, SE = 0.11, p = .136; t3: β = 0.01, SE = 0.13, p = .819). Accordingly, the predictive validity did not change over time (Δχ² = 3.02, Δdf = 2, p = .221). Overall, those aggregated analyses indicate support for Hypothesis 1c.

We conducted various robustness checks and additional analyses to either study the influence of our methodological choices (e.g., operationalizations or treatment of outliers) or of additional variables (e.g., different career trajectories). We report these analyses in the following.

Although we relied on mostly objective data, our findings may be influenced by a few unorthodox data points. Thus, we carefully screened the data for outliers (on the basis of Mahalanobis distances; e.g., Meade & Craig, 2012). Since this method cannot handle missing data, we additionally identified univariate outliers via boxplots to follow a conservative approach. Across the different operationalizations and time points, we detected 24 outliers for the coaches’ points per game, 39 outliers for the coaches’ goal ratio, and four outliers for the number of coached matches. We repeated our main analyses while excluding those outliers and found very similar results (see Table S2 in the supplemental file).

During the revision of the paper,¹⁰ we gathered additional data to capture further aspects of expert leadership beyond prior player performance (i.e., university degrees in relevant fields, such as sports sciences, and potential coaching licenses). Those variables have the potential to capture further facets of expertise and, thus, offer a more comprehensive test of TEL. Remarkably, only 47 coaches had a sports-related degree and former players (compared to coaches without that background) were significantly less likely to have obtained such a degree, χ²(1, N = 186) = 13.88, p < .001. Concerning the coaching license, only n = 19 coaches of our sample did not have a coaching license, whereas n = 18 were awarded with an A-license, n = 4 with a B-license, and n = 240 with a Pro-license (with the Pro-license being the most advanced license, which is nowadays a requirement in the Bundesliga). We found no data on coaching licenses for n = 127 coaches. Our analyses overall did not indicate a systematic link between sports-related degrees or coaching licenses and LP, neither initially nor over time (see Table 2, for correlations and Table S4 in the supplemental file). That is, only two of the 18 empirical links between sports-related degrees and LP were significant (see Table S4).

Some coaches started their coaching career as head coaches of a club in a first or second professional league within the UEFA, while others started their career in lower leagues or with youth teams. For the majority of the head coaches (n = 258; 63.39%), the head coach position in a first or second professional league within the UEFA (as a permanent or interim coach) was indeed their first coaching position, compared to n = 144 (35.38%) who first worked as a coach in a lower league or with youth teams, and n = 5 (1.23%) coaches with missing data. We tested the link between EP and LP for both groups of coaches (i.e., those who started their coaching career in a first or second professional league within the UEFA and those who started their coaching career in lower leagues or with youth teams). Due to the reduced sample size (i.e., only n = 101 of those 144 coaches had a previous player career), path models were not suitable. Hence, we conducted six multiple regression analyses using our different operationalizations of EP and LP for both coach groups. This approach allows for comparisons of the EP-LP link between both groups of coaches (i.e., those who started their coaching career in a first or second professional league within the UEFA and those who started their coaching career in lower leagues or with youth teams). Our analyses showed non-significant EP-LP links for both groups of coaches with one exception (i.e., the regression with the operationalizations player rating and the goal ratio at lower leagues indicated a negative relationship; for further information, see the supplemental file, Table S5).

As TEL proposed processes that should unfold within an organization over time (e.g., positive effects of hiring applicants with high potential), those effects cannot be tested conclusively with leaders who change teams relatively frequently (as in our study). When restricting our sample to coaches that did not change their club for three seasons, our sample was again too small to conduct path analyses (n = 77; see Kline, 2011). Thus, in an exploratory analysis, we computed partial correlations (i.e., controlling for prior team quality) and Steiger’s z-test (Diedenhofen & Musch, 2015; Steiger, 1980) to test whether the relationship between prior player and later coach performance differed over time (e.g., comparing the link between the average player rating and PPG at t1 and t2). The link between EP and LP increased descriptively over time for only two of the six combinations of operationalizations (i.e., average player rating with PPG, t1: r = -.25, p = .151; t2: r = .01, p = .974; t3: r = .21, p = .244; average player rating with goal ratio, t1: r = -.19, p = .296; t2: r = .04, p = .838; and t3: r = .15, p = .388). However, despite the descriptive increase the correlations did not change significantly over time (|zs|< 1.95, ps > .05).

With this study, we aim to initiate an exchange between different disciplines (i.e., personnel selection and management research) on the question of whether high-performing employees may later truly make successful leaders. To do so, we relied on the performance requirements perspective (Zaccaro, 2012), the follower-centric perspective (Uhl-Bien et al., 2014), and TEL (Goodall & Bäker, 2015) to derive our hypotheses: In particular, we specified hypotheses on the initial validity of performance-based promotion as well as on its validity over time. Due to different assumptions on mediating processes (such as performance requirements vs. perceived leader prototypicality vs. better strategic decision making) those perspectives led to contrasting hypotheses on the predictive validity of EP for LP. We provided the first empirical test on these contrasting hypotheses. Thereby, we systematically examined how the link between EP and LP unfolds over time (i.e., initially after a promotion and across consecutive seasons). Our study results (i.e., no systematic associations between EP and LP, neither initially nor over time) are consistent with the performance requirements perspective (see Zaccaro et al., 2018), but not with the follower-centric perspective or with TEL. Moreover, our findings are mostly in line with the conclusion of a recent review (Schleu & Hüffmeier, 2021).

The performance requirements perspective specifies the predictive validity of EP for later LP by comparing the performance requirements of the employee and the leader position. Due to the different tasks of employee and leader positions (cf. Bartram, 2005; Borman & Brush, 1993), employee performance requirements match only to a small degree with the performance requirements for the leader. Our research supports this general reasoning and the importance of this theoretical perspective when predicting the overall EP-LP link.

With an increasing overlap of performance requirements of EP and LP, the validity of performance-based promotion should increase. We tested job complexity (as one way to increase the overlap of performance requirements) of the employee position as a potential moderator of the relationship between EP and LP. Our results, however, indicated no support for the proposed moderation in our study context. To further clarify whether the theoretical rationale concerning an increasing overlap in the respective performance requirements is correct, further studies are needed to examine job complexity (and other proposed moderators) more comprehensively. According to the assumptions of the performance requirements perspective, complexity moderates the small link between EP and LP. Consequently, a comprehensive test might require a bigger sample size in the future. This might allow a more informed use of performance-based promotion in the future.

Our results did not support the follower-centric perspective concerning performance-based promotion in the context of our study. Future research could clarify under which conditions these findings hold (within and beyond sports) and thereby also improve the current understanding of the follower-centric perspective. In particular, it might be promising in more conventional organizational contexts, where subjective performance evaluation is more common, to consider several sources before and after a promotion for a comprehensive analysis of the perceived prototypicality, EP, and LP (while avoiding the same rater bias; Hoyt, 2000). Potentially, such biases could inflate the perceived validity of performance-based promotion and therefore ensure the ongoing prevalence of this strategy in practice.

Our results provided no support for TEL (Goodall & Bäker, 2015) in the context of our study (i.e., professional soccer), as the analyses indicated overall no systematic link between EP and LP. As the processes proposed by TEL should unfold within an organization over time (e.g., positive effects of hiring applicants with high potential), it is conceivable that it may not be possible to test these effects conclusively with leaders who change teams relatively frequently (as in our study). When restricting our sample to coaches who did not change their club over the course of the least three seasons (in a post-hoc analysis), our findings were inconsistent concerning the hypothesized effects over time in this substantially smaller subsample (i.e., two of the six analyses indicated a descriptive increase of the EP-LP link over time). To adequately test the proposed effects of TEL in the future, a rather stable study context (i.e., avoiding power problems) or shorter time lags (i.e., coach performance per week or match day) might be beneficial, as the coaches oftentimes worked only one season for a club.

As the majority of prior studies on performance-based promotion reported negative, non-significant, and mixed findings (for an overview, see Schleu & Hüffmeier, 2021) and our results did not indicate a systematic link between EP and LP, the empirical evidence questions the general validity of the logic underlying performance-based promotion. Some variability in the results of extant studies could potentially be explained by the varying importance of the different theoretical perspectives for explaining the EP-LP link across different contexts. Hence, it might be promising to theorize on boundary conditions for the different theoretical perspectives to understand when, for instance, the follower-centric perspective is equally or even more important than the performance requirement perspective.