Manager–employee relationships

A manager and an individual contributor (a direct report with no reports of their own) form a working relationship whose quality is associated with ratings, trust and delegation. Research describes that quality mainly through leader-member exchange (LMX), a dyadic, relationship-based approach to leadership (Wikipedia contributors, n.d.a), and describes the employee’s efforts to shape it as upward influence, feedback seeking, voice and impression management; practitioners call the same efforts “managing up”. The central dispute is how much of the relationship is the employee’s to improve: practitioner advice places the work mostly on the employee, while the strongest research findings place most of the explained variance on the manager. Much of the evidence on what employees can do is thin, correlational, or rests on opinion.

History

No history of the field is available yet (the origins of LMX theory, of feedback-seeking research or of the managing-up literature). The Harvard Business Review article "Managing Your Boss" by Gabarro and Kotter was first published in 1980 and republished in 2005 (Gabarro & Kotter, 2005); a fuller account of how these literatures developed still needs research.

Core concepts

Terms for the same activity

Several literatures name the employee’s side of the relationship differently.

• Managing up. Gabarro and Kotter define managing your boss as consciously working with the superior to get the best results for the employee, the boss and the company, and distinguish it from political manoeuvring and flattery (Gabarro & Kotter, 2005) (Gabarro & Kotter, 2005). Culture Amp frames it as building an effective relationship with the manager, a partnership (Boogaard, 2025).
• Upward influence. A textbook defines it as the ability to influence the boss and those above, and lists tactics derived from Kipnis’s work, such as rational persuasion: using facts, data and logical arguments (Saylor Academy, n.d.).
• Leader-member exchange. LMX describes the quality of the relationship between a leader and one follower. A high-quality relationship (the “in-group”) is marked by mutual trust, respect and obligation, and although the leader dominates, the follower also helps to create it (Wikipedia contributors, n.d.a) (Wikipedia contributors, n.d.a).
• Followership. A role-based approach equates followers with people of low formal power, that is, direct reports. A process approach instead defines followers as those being influenced toward shared goals, whatever their rank, so that leadership and followership are one social influence process (Bastardoz & Adriaensen, 2023). Bastardoz and Adriaensen argue that a direct report’s upward leading and the manager’s following emerge together (Bastardoz & Adriaensen, 2023).

What predicts relationship quality

A meta-analysis of 247 studies (290 samples) tested 21 antecedents of LMX quality and found that leader variables explained the most variance (Dulebohn et al., 2012). Follower characteristics also correlate with LMX: competence, agreeableness, conscientiousness, extraversion, internal locus of control and positive affectivity correlate positively, and negative affectivity and neuroticism negatively. These are correlations rather than shown causes, and they are traits more than behaviours (Wikipedia contributors, n.d.a). A summary of the same research reports that interpersonal variables such as perceived similarity, liking, self-promotion and the leader’s trust correlate positively with LMX, with assertiveness an exception (Wikipedia contributors, n.d.a).

Feedback seeking

In a sample of 209 supervisor-subordinate dyads in China, subordinates’ feedback seeking was positively but weakly related to LMX (beta .14, R² .02), and more strongly when supervisors read it as task-driven than as impression-driven (Lam, 2006). The link held only when supervisors saw weak impression-management motives (simple slope .41); with strong impression-management attribution it was not significant (Lam, 2006). The thesis author states that no earlier study had empirically tested whether subordinate feedback seeking relates to LMX (Lam, 2006). A published study reports the same pattern: feedback seeking was linked to higher LMX and performance only when supervisors attributed it mainly to performance enhancement rather than impression management (Lam et al., 2007).

The direction of cause is not settled. In a cross-sectional survey of 134 Korean civil engineering employees, higher LMX was linked to a preference for direct feedback-seeking strategies, but not to indirect or third-party ones, which suggests that LMX shapes feedback seeking (Lee et al., 2007). In the other direction, when supervisors seek feedback from subordinates (259 subordinates in 52 teams), subordinates tend to respond with more upward ingratiation rather than sincere feedback (Mertens et al., 2024).

Voice and taking charge

In two paired employee-supervisor studies in Taiwan, employee voice and taking charge were linked to LMX through the leader’s perception of the employee as constructive, and the link was stronger for employees with better past performance (Shih & Nguyen, 2023). Managers also respond to how often an employee speaks up. They rated employees with moderate voice frequency more favourably than employees who rarely or very frequently spoke up, and the inverted-U for prohibitive voice (pointing out problems) depended on LMX quality (Huang et al., 2018). The authors conclude that raising problems with a manager may be less risky than assumed: a moderate frequency can project a positive image, and only very frequent prohibitive voice tends to draw unfavourable responses (Huang et al., 2018).

Whether employees speak up depends on the manager as well. A systematic review of 26 studies concludes that employees speak up to leaders more when the leader signals openness, responsiveness and legitimacy for upward communication (Edon et al., 2026). In a survey of healthcare employees, empowerment through high-quality LMX and a low “bottom-line mentality” reduced the tendency to withhold or soften bad news to supervisors, which the authors call the Hierarchical MUM Effect (Scrimpshire et al., 2021).

Communication frequency and format

How often an employee communicates with the supervisor changes what LMX does for performance ratings. In a private-sector sample (N=188), replicated in a public-sector sample (N=153), frequent communicators were rated better when LMX was high and worse when LMX was low (Kacmar et al., 2003). Practitioner guides add advice on form: Osmani recommends that senior individual contributors communicate status, risks and blockers early, in the manager’s preferred format, connect their work to business goals, and state their career goals and the support they need (Osmani, 2024). Kaplan-Moss advises that status updates to executives be very short and focus on what the audience needs to know, so that silence is read as “nothing to flag” (Kaplan-Moss, 2021). Another blog advises tailoring messages to the manager’s communication style and bringing proposed solutions with problems, and gives no evidence for either (Jedrzejewski, 2024).

How managers come to trust employees

Research on trust in organisations has mostly studied subordinates’ trust in supervisors; supervisors’ trust in subordinates is understudied (Effects of subordinate's job performance, 2009). A review of 73 trust studies found little attention to subordinate attributes as antecedents of trust, so what employees do to earn trust is thinly documented (Nienaber et al., 2015). The relative importance of ability, benevolence and integrity in predicting trust differs depending on whether the trustee is a supervisor, a subordinate or a peer (Knoll & Gill, 2011). In a doctoral thesis on leaders’ trust in followers, these general antecedents predicted leader trust more strongly than specific ones (availability, receptivity) (Zheng, 2018).

A survey of supervisors found that supervisors trust high-performing subordinates generally. When task performance is low, helpful behaviour directed at individual colleagues is strongly and positively related to supervisor trust, and the supervisor’s perception of the subordinate’s altruistic motive fully mediated that link: the same helpful act builds trust only if it is read as altruistic (Effects of subordinate's job performance, 2009). Practitioner sources add that trust is built as a repeated pattern and eroded most by the manager hearing of a problem from someone else, and that a mistake the employee discloses is forgivable while one the manager discovers is not (Buteau, 2026a). A career guide for programmers defines reliability as seeing assigned tasks through, and says that a report who withholds information invites micromanagement (Holloway, 2023). These practitioner claims rest on opinion rather than tested evidence.

Questions, controversies, and new developments

Partnership or politics?

Practitioner sources agree that managing up is honest work toward a shared outcome. Buteau separates it from politics: managing up is an operational skill that includes candour and pushback, while politics is telling the manager what they want to hear or hiding bad news (Buteau, 2026b). A trade-magazine piece quotes a coach who says dealing honestly and upfront with a manager is not sucking up, whereas political manoeuvring makes managing up ineffective (Chartered Accountants Worldwide, n.d.).

The research complicates that line, because the tactics practitioners warn against do work to some degree. A meta-analysis found ingratiation to have a small positive effect on performance evaluations and a larger effect on liking, moderated by how transparent the ingratiation appears and by its direction (Gordon, 1996). A later meta-analysis found other-focused impression-management tactics (ingratiation) significantly related to performance ratings (r_c = .25), while self-focused tactics (self-promotion) were not (r_c = .18, not significant); the outcome there is ratings, not LMX (Peck & Levashina, 2017). Peck and Levashina argue that self-promotion toward a supervisor can backfire, because supervisors know the employee’s competence and may not believe discrepant claims, and repeated self-promotion risks seeming arrogant (Peck & Levashina, 2017). In a lab study, repeated ingratiation raised supervisor performance ratings over time while repeated apologies lowered them, and the effects of apologies and justifications were stronger early in the relationship (Bolino et al., 2014). For newcomers, a supervisor’s early perception of ingratiation lowered initial LMX, an effect buffered by the supervisor’s self-enhancement tendency (Wang & Liu, 2020). The line between politics and managing up therefore seems to turn on how the manager perceives the motive, not on whether the tactic is effective.

Which tactics work?

A meta-analysis of 11 influence tactics found rational persuasion the only tactic with stable positive relationships to task and relationship outcomes regardless of moderators, and pressure negatively related to outcomes (Lee et al., 2017). The meta-analysis covers influence in all directions, not only upward. Across cultures, upward influence strategy patterns only partly match what cross-cultural values research would predict, in an exploratory study of six cultures (Ralston et al., 2001). In a master’s thesis of more than 200 Taiwanese supervisor-subordinate dyads, the supervisor’s interpersonal trust mediated the link between subordinates’ upward influence strategies and delegation, and power-distance orientation moderated the effect of soft versus hard strategies on trust (Cheng, 2016).

Whose responsibility is the relationship?

Practitioner sources frame managing up as work the employee does, and give no evidence on how much of the relationship it can change (Buteau, 2026b; Chartered Accountants Worldwide, n.d.). The research points the other way: leader variables explained the most variance in LMX quality (Dulebohn et al., 2012), and in a Finnish interview study of remote work (81 interviews), leaders put in more effort than subordinates to maintain a good feedback environment, though the authors call for more proactive subordinates (Jansson & Kangas, 2024). The two positions are not strictly opposed, since LMX is co-created (Wikipedia contributors, n.d.a), but the research gives no basis for weighting the employee’s share above the manager’s.

Strength of the evidence

The behavioural findings are small or conditional. The feedback-seeking link to LMX explains about 2 percent of variance in the largest dyad sample reported here (Lam, 2006), and its effect disappears when the manager suspects impression management. Most studies of voice, feedback seeking and trust are correlational, and the direction of cause is open. Evidence that helpful behaviour earns trust comes from one survey, and it depends on how the supervisor interprets the motive (Effects of subordinate's job performance, 2009).

Broader connections

Coordination, shared understanding and agreement

The employee’s work of keeping a manager informed is a case of the wider problem of knowing who needs to know what, treated in Cross-team coordination. What a manager and a report each hold in mind about a task is the subject of Common ground. A tendency to tell the manager what they want to hear, which practitioners call politics, has a counterpart in language models, covered in Sycophancy in language models.

Directing an AI agent

Practitioner essays compare delegating to an AI agent with managing an employee. Koh argues that observation, process and graduated-trust practices from human delegation transfer to AI, but that the social contract (reputation and career consequences) does not, so architecture must substitute for accountability (Koh, 2026). Spinellis says delegation to AI follows the same logic as delegation to people (goal, authority, monitoring, accountability) but differs in speed, convincing errors, lack of memory and unlikeliness to object, and that failures such as sycophantic agreement are harder to spot than human evasion or inexperience (Spinellis, 2026). A vendor blog argues that treating AI like an intern limits it, that delegating ownership with check-ins and escalation thresholds works better, and that agents given persistent context do not forget coaching as people do (W. & Scout Team, 2026). All three are essays.

Some experimental work bears on the agent side of the pairing. In a pre-registered experiment, first-person uncertainty phrasing by a language model lowered users’ confidence in the system and their agreement with its answers while raising their accuracy; general-perspective phrasing had weaker, non-significant effects (Kim et al., 2024). The hedges improved accuracy by reducing overreliance on wrong answers, so lower trust coincided with better outcomes (Kim et al., 2024). In a study of a workflow agent, users still wanted control at points of uncertainty, and the agent’s reasoning was exposed in a dashboard for evaluation (Long et al., 2025).

Asking questions is the closest analogue to feedback seeking. In a user study of document generation, follow-up questions improved document preference and user experience, and users valued open-ended or insightful questions over plain information requests (Tix, 2024). In an online product-search experiment, users answered about 11 to 21 clarifying questions on average and some stopped because of fatigue or irrelevant questions (Zou et al., 2020). Neither study tests how users attribute motives to the agent.

Further reading

• Gabarro and Kotter, “Managing your boss” (Gabarro & Kotter, 2005): the classic practitioner statement of managing up as a deliberate partnership.
• Dulebohn and colleagues, a meta-analysis of antecedents and consequences of LMX (Dulebohn et al., 2012): the main evidence that leader variables explain most variance in LMX.
• Lam, Huang and Snape, feedback seeking and LMX (Lam et al., 2007): feedback seeking was linked to higher LMX and performance only when supervisors read it as performance enhancement rather than impression management.
• Huang and colleagues, promotive and prohibitive voice (Huang et al., 2018): voice frequency and managers’ responses.
• Peck and Levashina, impression management meta-analysis (Peck & Levashina, 2017): other-focused versus self-focused tactics and the risk of self-promotion.
• Spinellis, “Barings, Boeing, and bots” (Spinellis, 2026): a short essay on where delegating to AI matches and departs from delegating to people.