Transformational leadership and leader-member exchange: a critical review

TFL is primarily a leader-focused theory, and majority of studies highlight four core behaviors of transformational leaders: 1) idealized influence, demonstrating charismatic behaviors focused on values and purpose; 2) inspirational motivation, motivating followers with a compelling vision and ambitious goals; 3) intellectual stimulation, encouraging critical thinking and questioning assumptions; and 4) individualized consideration, providing personalized support (Antonakis et al., 2003).

These behaviors give meaning and reasons to employees (Ng, 2017), which ultimately lead to positive outcomes in the organization (for a review, see Wang et al., 2011). LMX, on the other hand, is a relationship-based theory in which leaders and followers are not the principal interests, and instead, the leader’s effectiveness is associated with the quality of the dyadic relationship with followers (Graen & Uhl-Bien, 1995). A high-quality LMX involves trust, respect and loyalty, which enhance interpersonal interaction and dyadic influence, and consequently lead to a range of individual and organizational outcomes (for a review, see Martin et al., 2010).

Conceptually, however, there is notable evidence in the literature showing that TFL incorporates elements of high-quality LMX and vice versa. On the one hand, the literature suggests that TFL’s underlying concept involves leaders’ capability to develop a high-quality relationship with followers (Howell & Hall-Merenda, 1999), which may explain the reliance of TFL research on social exchange theory (Gottfredson & Aguinis, 2017) and TFL’s shared common characteristics with high-quality LMX (e.g. communication quality: Graen et al., 2010). In other words, TFL effectiveness seems to depend on the quality of the leader’s relationships with followers. In fact, some scholars argue that TFL, at its core, is about building high-quality relationships with followers (Deluga, 1992). The different relational tactics used by transformational leaders are based on the quality of exchange relationships (Gottfredson & Aguinis, 2017), and transformational leaders influence their followers by arousing feelings of personal identification (Kark et al., 2003). Some scholars highlight how the different sub-dimensions of TFL exemplify relational-based characteristics. For example, Seltzer & Bass (1990) suggest that idealized influence may have a dyadic nature. Similarly, Hollander (1993) confers that individual consideration and intellectual stimulation have an exchange-based nature. A transformational leader is aware of each follower’s developmental level and adjusts their behavior to meet those individual needs (Avolio & Bass, 1995); however, individualized consideration seems to satisfy followers’ needs for belonging, acceptance and human interaction, all of which are characteristics of exchange relationships (Basu & Green, 1997; Saboe et al., 2015). According to Kent & Chelladurai (2001), “the definition for individual consideration as treating subordinates ‘differently according to each subordinate’s needs and capabilities’ (Bass, 1985, p. 82) is reflective of LMX” (p. 152).

On the other hand, evidence in the literature suggests that high-quality LMX seems to be more transformational than an exchange-based relationship (Gerstner & Day, 1997). The high-quality exchange reported in LMX does not measure the exact nature of an exchange, and instead, it seems to be based on the exchange of values (Graen & Uhl-Bein, 1995), which is similar to the idealized influence sub-dimension of TFL where sharing values and sense of purpose are emphasized (Gillespie & Mann, 2004). In fact, some researchers discuss that while LMX may initially start as an exchange-based relationship, it evolves over time into a more transformational relationship (Hughes et al., 2010). As Krishnan (2005) noted, in high-quality LMX, leaders do not explicitly ask followers for higher performance or harder work, and it seems high-quality LMX is more transformational in nature. Additionally, any dyadic relationship inherently reflects the existence of two actors (e.g. leader and followers), without which the relationship would not exist (Dansereau et al., 1995). As noted by House & Aditya (1997, p. 431), “it is almost tautological to say that good or effective leadership consists in part of good relationships between leaders and followers.” This suggests that although LMX focuses on the relationship between leader and follower, such a relationship requires both parties and cannot exist independently of them. Furthermore, some specifications of high-quality LMX, including trust, mutual respect and openness, are more indicative of leader behaviors than the relationship exchange (Antonakis et al., 2014), suggesting that these characteristics may be better understood as leadership behaviors rather than relational process. Another indication of similarity between TFL and high-quality LMX is intervention training to increase the quality of exchange relationships, but these training activities seem more reflective of TFL behaviors in seeking to understand employees’ viewpoints. These training interventions include talking with followers about their problems, listening to their concerns, not imposing their perspective as the leader and discussing mutual expectations and experiences (Gerstner & Day, 1997). In sum, the literature suggests conceptual similarities between TFL and high-quality LMX.

Two constructs are empirically distinct if they are not highly correlated and they do not show similar patterns of correlations with other variables, known as nomological networks (Le et al., 2010). The literature, however, reports a wide range of correlations between TFL and LMX. For example, Zhang, Jia & Gu (2012) reported a correlation of r = 0.24 between TFL and LMX, p < 0.001, whereas Shin, Seo, Shapiro & Taylor (2015) reported r = 0.90, p < 0.01. Such a wide range of correlations does not provide a definite answer regarding the magnitude of the correlation between TFL and LMX, which in turn complicates the interpretation of their nomological networks.

There are several plausible reasons why the extant research reports such a wide range of correlations. One explanation may be the considerable inconsistencies in both the number of items and sub-dimensions of scales measuring TFL and LMX. Each scale’s unique characteristics affect the variance of the true score (Le et al., 2010), which ultimately affect studies’ results. The literature shows that the number of items in various TFL scales could range from as low as five (e.g. Bettencourt, 2004) to as high as 47 (e.g. Gupta & Krishnan, 2004). Similarly, TFL scales’ sub-dimension ranges from as low as one (e.g. Pundt, 2015) to as high as six sub-dimensions (e.g. Wang, Law, Hackett, Wang & Chen, 2005). In a similar vein, different LMX scales use varying numbers of items (e.g. 2, 4, 5, 7, 10, 12 and 16 items) and different dimensions (unidimensional vs and multi-dimensional) as noted by Graen & Uhl-Bien (1995). Another explanation for this wide correlational range could be the nature of individual studies, which often have limited sample sizes and are subject to biases (e.g. measurement error) that may affect their true results (Hunter & Schmidt, 2004). For example, Basu & Green (1997) found a large and significant correlation between TFL and LMX, and discussed that methodological artifacts may be one possible reason for such results.

To overcome the limitations of individual studies, researchers use the meta-analytic method, which integrates the findings of multiple individual studies and benefits from a larger sample size and greater statistical power. Furthermore, a meta-analysis can address research artifacts associated with individual studies (Hunter & Schmidt, 2004) and rule out the possibility of their effects on the results. Most importantly, meta-analyses combine findings across different definitions and measurement scales, which is crucial for the purpose of our review paper to synthesize and interpret the existing body of research. These advantages enable meta-analytic studies to deliver more generalizable and conclusive results (Hunter & Schmidt, 2004). As meta-analysis is a quantitative method, reviewing meta-analytic studies instead of individual studies could offer more reliable insights into the empirical similarity between TFL and LMX. This novel review method is called umbrella review (Grant & Booth, 2009). Since meta-analysis is a powerful tool that mitigates the discussed limitations of individual studies, an umbrella review can provide an even more robust answer to the research question with inconclusive findings (Papatheodorou, 2019), which serves our purpose of examining the effect of conceptualizations of TFL and LMX on empirical findings by synthesizing results of previous meta-analyses and providing a more robust conclusion.

While the umbrella review has received great attention in other fields like health care, it is still a relatively new and underexplored method in the management field, likely due to the previously insufficient number of review studies required for such a method. However, the recent rapid growth of review paper publications in the management field over the past decade (Kraus et al., 2024; Kraus et al., 2020) makes it well-suited for using the umbrella review.

To conduct an umbrella review, we reviewed previous meta-analyses to find meta-analytic correlations between TFL and LMX and compare their nomological networks. Similar to Hoch et al.’s (2018) approach, we checked the nomological network with behavioral outcomes (i.e. performance, citizenship behavior and deviance), attitudinal outcomes (i.e. satisfaction; commitment) and relational perception (i.e. trust). Furthermore, as one of the sub-dimensions of TFL is inspirational motivation, we also included motivation as an additional attitudinal outcome.

We conducted searches in databases, including Scopus, ScienceDirect and PsycINFO up until December 2024. We used keywords in our search such as meta-analysis, meta-analytic, review, leader, leadership, leadership behavior, leader-member exchange, LMX, dyad relationship, transformational leadership, individual consideration, intellectual stimulation, idealized influence and inspirational motivation. Furthermore, we checked reference lists and conducted manual searches in Google Scholar (Greenhalgh & Peacock, 2005). We selected English-language studies that reported meta-analytic Pearson correlation coefficients as effect sizes among our variables of interest across different theoretical frameworks. After removing duplicates, our criteria resulted in 34 journal articles and two book chapters (these meta-analytic studies are marked with asterisks in the References). Overall, the meta-analyses used in this review have good quality and scientific rigor. Among the 34 selected journal articles for this review, 20 are published in A*-level, eight in A-level, and five in B-level journals, according to the Australian Business Deans Council (ABDC) list ^[2]. One paper is not in the ABDC list, but it is Q2 ranked (the second highest quartile) in psychology by Scientific Journal Rankings (SJR) ^[3]. Both book chapters are from volumes on leadership edited by renowned scholars in the leadership field (see Dumdum et al., 2002 & Hackett et al., 2003).

Finally, out of these 36 meta-analyses, there is one meta-analysis before 2000, nine meta-analyses from 2000 to 2010, 19 meta-analyses from 2010 to 2020 and seven meta-analyses from 2020 to December 2024. This trend shows the growing attention to meta-analyses in the leadership field and supports our approach to use the umbrella review method. In total, these selected meta-analyses provided us with 122 meta-analytic correlations to check the two discussed steps for empirical similarities.

The findings of our review show that meta-analytic correlations between TFL and LMX range from 0.69 to 0.77 (Table 1). Schober, Boer & Schwarte (2018) categorized correlation coefficients in the range of 0.70–0.89 as “strong correlation” (p. 1765). In the leadership context, Banks, Gooty, Ross, Williams & Harrington (2018) suggest that a correlation above 0.70 between two leadership constructs is “alarming” (p. 237). The results suggest that the correlations between TFL and LMX are alarmingly high. Furthermore, the sample sizes are sufficiently large to conclude that correlations are reliable estimates of population.

Table 2 presents the nomological network correlations of TFL and LMX. The findings show that the patterns of correlations are different. LMX shows stronger correlations with employee performance, citizenship behaviors, overall job satisfaction, and commitment. On the other hand, TFL is more strongly associated with motivation, satisfaction with the leader, and reduced deviance. Such differentiated associations with theoretically relevant outcomes suggest evidence of distinctiveness and rule out the possibility of redundancy (Riketta, 2005). Furthermore, because these patterns are based on meta-analytic effect sizes aggregated across multiple studies, they provide relatively consistent pictures of how each leadership construct relates to key outcome variables. Our findings also show that only one meta-analysis has examined LMX’s effect on team performance, making it infeasible to compare with TFL’s effect. Regarding trust, results are mixed.

The results show alarmingly high correlations between TFL and LMX. However, the findings show different correlational patterns between TFL, LMX, and their shared nomological network. While high correlations suggest a significant overlap between TFL and high-quality LMX, different correlational patterns provide support for their distinction and eliminate the possibility of redundancy. Furthermore, these results provide a few other noteworthy insights about TFL and LMX. First, prior research has reported different correlations across studies between TFL, LMX, and their shared nomological networks, leaving it unclear which construct is more strongly associated with them. The results of this paper address this gap by providing comparative evidence of their relational strengths and clarifying their unique roles in the leadership literature. Second, this review presents ranges of meta-analytic correlations and provides a clearer picture of the relationship between TFL and high-quality LMX. The findings suggest that using one pair of correlations – either individual or meta-analytic – to check nomological similarity between TFL and LMX may fail to capture the variability across different studies and result in misleading conclusions by overlooking the broader pattern of relationships, a potential limitation in studies that use such an approach. Continuing the discussion, we next discuss potential explanations for the high correlation and the observed patterns of relationships.

The alarmingly high correlations between TFL and LMX represent an excessive degree of overlap between them, which may support the previously discussed conceptual similarity between TFL and high-quality LMX and highlight the need for more apparent theoretical differentiation. The quality of research in a discipline depends on the quality of its constructs (Cole et al., 2012), and the discussed conceptual similarities between TFL and high-quality LMX could blur definitional clarity and weaken construct validity (Schwab, 1980).

Beyond the discussed conceptual overlap, methodological issues may provide other alternative explanations for the observed overlap. First, research studies discussed the similarity of TFL scales to LMX scales (Kent & Chelladurai, 2001). Similarly, some items in LMX scales seem to be more compatible with TFL conceptualization (Gottfredson & Aguinis, 2017). The similarity of scales and failure to measure the construct as conceptualized may lead to biases that inflate the correlation between TFL and LMX, including confounding bias and implicit theory and illusionary correlation bias (Gottfredson & Aguinis, 2017; Martinko et al., 2014; Podsakoff et al., 2003). Second, scholars have already warned about single-source bias and its effect; however, we believe that single-source bias is even more problematic in studies with both TFL and LMX, in which followers rate both constructs. Like any other perception, follower’s perception of leadership is subjective (Brown & Keeping, 2005; Yammarino & Dubinsky, 1994) and may depend on the quality of their relationship. Thus, it is not surprising that those followers with high-quality relationships, who receive greater attention from their leader and better access to resources, also rate their leader as more transformational. Moreover, followers’ perceptions may be biased by their feelings toward the leader, a phenomenon called affective driven evaluation (Brown & Keeping, 2005; Hall & Lord, 1995). In this situation, liking is a source of variance in raters’ evaluation (Brown & Keeping, 2005), which could affect subordinates’ ratings about the extent to which their leader is transformational and lead to an inflated correlation between TFL and LMX.

These conceptual and operational issues that appear to inflate the correlation between TFL and LMX reiterate existing critiques on these constructs. A notable recent critique is by Gottfredson, Wright & Heaphy (2020), who conducted a conceptual critique of LMX through analyses of 42 individual studies published between 2014 and 2019. Our review complements and extends their critiques by providing empirical findings to the discussed critics. The empirical insight provided by the current study is based on a broader data set of 36 meta-analytic studies from 1997 to 2023, which enhances the scope and generalizability of the points raised in the critique. Despite the valid concerns that may contribute to the inflated correlation between TFL and LMX, our findings suggest that TFL and LMX differ in the pattern of their relationships with their shared nomological network. This contrasts with Gottfredson et al.’s (2020) assertion that the measurement of TFL and LMX is indistinguishable. Such a claim implies that TFL and LMX correlate similarly with other variables, which our review does not support. While conceptual and operational imperfections in TFL and LMX warrant researchers’ attention, they do not fully obscure the distinction between the two constructs. Instead, our results indicate that they still capture unique aspects of leadership, which support their theoretical and empirical distinction. Our findings highlight the importance of integrating empirical analysis with conceptual critique, as distinct yet complementary lines of evidence, in evaluating the controversial issues in the literature.

The results show that LMX has stronger associations with employees’ performance, citizenship behaviors, overall satisfaction, and commitment, reemphasizing the role of quality relationships. These results are also compatible with models in which LMX mediates the relationship between TFL and workplace outcomes (e.g. Wang et al., 2005). In such mediation models, LMX has a direct effect and TFL has an indirect effect, which aligns with our findings that LMX shows stronger correlations with specific workplace outcomes. These models also illustrate that high-quality LMX is shaped and supported, in part, by TFL behaviors. As noted by Wang et al. (2005, p. 423), “transformational leadership builds and nourishes high-quality LMX,” highlighting the role of TFL in high-quality LMX. However, the facilitation of high-quality LMX by TFL in the mediation model should not be interpreted as redundancy; instead, it reflects an integrative perspective in leadership studies (Avolio et al., 2013), and the causal sequence, where TFL leads to high-quality LMX, is a defining feature of the mediation model (MacKinnon et al., 2007).

The findings of this review, however, add two caveats to these mediation models. First, while there should be a correlation between a predictor variable and a mediator variable in any mediation model, a high correlation between them could complicate interpreting the unique role of the mediator, which is known as multicollinearity (see Shrestha, 2020). Considering the findings of this review, a correlation above 0.70 between TFL and LMX means that more than half of the variance (r² ≥ 49%) in LMX may be statistically explained by variation in TFL, making it difficult to distinguish the unique effects of LMX on outcome variables. Second, the multicollinearity causes the direction of mediation to become more unclear than typically acknowledged. In other words, the reverse relationship, where TFL is the mediator, could also be plausible. In this view, a high-quality LMX relationship is a precursor to effective TFL. This possibility is consistent with an alternative model developed by Boer, Deinert, Homan & Voelpel (2016), in which TFL mediates the relationship between LMX and workplace outcomes. This possibility may be particularly true for outcomes where TFL shows stronger correlations in our results (i.e. motivation; satisfaction with the leader; reduced deviance).

Our results that show TFL has a stronger relationship with motivation seems to align with TFL’s conceptualization. The earliest definitions of TFL highlighted the role of motivation. Burns (1978) defined TFL as the process in which “leader and followers raise one another to higher levels of morality and motivation” (p. 20). Later, Bass (1985) defined transformational leaders as leaders who motivate followers to perform beyond expectations. The capacity of transformational leaders to motivate followers seems to have been downplayed in the research over time. However, the inspirational motivation sub-dimension of TFL may capture this type of leader behavior. As discussed earlier, some scholars highlight the similarities between the other three sub-dimensions of TFL and high-quality LMX, but inspirational motivation seems to distinguish TFL from an exchange-based relationship. Some empirical evidence may support the uniqueness of the motivational sub-dimension of TFL. For example, there are studies that report inspirational motivation – among four sub-dimensions of TFL – has the weakest effect or no effect in predicting LMX (e.g. Li & Hung, 2009; Zacher et al., 2013). Furthermore, research studies suggest that inspirational motivation adds affective quality to “arouse followers’ emotions and motivation” (Rafferty & Griffin, 2004, p. 332), and this affective component may explain why followers are more satisfied with transformational leaders.

Our results also show that TFL has a stronger relationship with reduced deviant and counterproductive behaviors, which are negative behaviors that may harm the organization and employees (Capitano & Cunningham, 2018). As noted by Colquitt et al. (2013), the social exchange framework does not typically consider avoiding negative behaviors as a means of reciprocating for the benefits received in an exchange relationship, explaining why LMX does not show a stronger correlation with reduced deviant behaviors. Finally, trust is a key component of both the idealized influence sub-dimension of TFL (Legood et al., 2021) and the exchange relationship in LMX (Colquitt et al., 2013), which may explain why neither TFL nor LMX shows a stronger relationship with trust.

Due to the existing evidence suggesting potential conceptual similarities between TFL and high-quality LMX, scholars are encouraged to refine the literature by providing clearer definitions of both constructs. From an operational perspective, efforts should be made to decrease similarities between TFL and LMX scales and also address inconsistencies across different scales (e.g. inconsistencies in sub-dimensions of TFL). Particularly, researchers are encouraged to consider the original value of motivation in the TFL scale items to improve its uniqueness. Furthermore, we reiterate previous recommendations to use different rating sources and measure TFL and LMX across time to partially remedy the discriminant validity concern between them (Podsakoff et al., 2003).

The study by Boer et al. (2016) appears to be the only one that has tested models in which TFL mediates the relationship between LMX and workplace outcomes. We encourage researchers to further explore models where TFL is the mediator, particularly for workplace outcomes that TFL shows stronger correlations in our results (i.e. motivation, satisfaction with leader, reduced deviance). Our review also shows that one meta-analysis has examined the effect of LMX on team performance, which may be due to the limited number of primary studies in this area and underscores the need for further research. Finally, we encourage management researchers to take advantage of the benefits of the umbrella review method. In particular, the umbrella review method could facilitate future comparative reviews in the leadership field, where new constructs emerge faster than in any other organizational behavior topic (Banks et al., 2018). We hope this paper makes management scholars more familiar with this method and encourages its adoption in the field.

Although our findings show that TFL and LMX are conceptually and/or empirically conflated, they also reveal that these two constructs exhibit distinct patterns of correlations within their shared nomological network. These findings provide clearer guidance for practicing managers to consider how each construct relates to specific workplace outcomes. As Lyubykh, Turner, Hershcovis & Deng (2022) noted, leadership effectiveness varies according to the particular criterion in focus, and thus, leaders must adapt their leadership focus based on the circumstances (DeRue et al., 2010). For example, our findings suggest that transformational leaders are more apt to reduce workplace deviance, implying that leaders who take the reins after damage to an organization’s reputation would benefit more from adopting TFL. Furthermore, the findings suggest implications for organizations in selecting and assigning leaders to specific roles or organizational contexts. For example, organizations undergoing change initiatives may benefit from transformational leaders, who can better motivate employees to embrace change (Gilley et al., 2009). By contrast, LMX seems particularly effective for promoting employee performance, citizenship, commitment and overall job satisfaction. For example, if employee job satisfaction is the criterion for leadership effectiveness in an organization (DeRue et al., 2011), LMX is likely to be more effective. Our study’s findings suggest that effective leadership requires broadening the focus to determine whether a behavioral or relational emphasis better guides practices to elicit desired employee responses.

The different nomological networks of TFL and LMX may reflect true conceptual and/or operational distinctions between the constructs and their relationships to the respective outcomes. However, there is a possibility that the observed effects are influenced by the endogeneity bias, particularly given that most leadership studies do not check the endogeneity (Gottfredson et al., 2020). Thus, while the results provide insight into the contributions of TFL and LMX in explaining different outcomes, they should be interpreted with appropriate caution and contextual understanding (Giarratana, 2024), and the findings should be a guiding framework rather than definitive causal evidence.

Introducing new leadership constructs similar to existing ones may obscure their distinct value and associated outcomes (Day, 2014). In this paper, the umbrella review method was adopted to provide a thorough understanding of the unique contributions of TFL and LMX. The umbrella review findings supported their distinction, but at the same time, the findings provided more robust evidence of significant overlap between TFL and high-quality LMX and supported the critiques in the literature about the potential conceptual and operational issues in TFL and LMX studies that require further attention. We hope this review, with both conceptual and empirical insights, offers an integrative and meaningful contribution to advancing the field.