The leadership cascade: manager behavior flows downhill

When a senior leader ignores employee feedback, the cost is not contained to their own team. It echoes down the org chart. The managers below them watch what their boss does, copy it, and pass it on to the people they manage. Behavior flows downhill.

We set out to measure that flow directly. Using 180 days of feedback data from 588 managers across 39 organizations on the Happily platform, we asked one question: do managers learn their management habits from their own managers? We used a specific, observable behavior as the test case — whether a manager replies to the text feedback their direct reports give them.

The answer is yes, and the effect is larger than we expected. A manager whose own manager replies to feedback is 2.4x more likely to reply to their own team's feedback. The probability of being a replier jumps from 49% when the boss does not reply to 81% when the boss does — an absolute lift of 32 percentage points.

How we measured the cascade

The behavior under study is the feedback reply. On Happily, a direct report can leave text feedback for their manager. The manager can reply to it or leave it unanswered. A manager's reply rate is the proportion of text feedback from their reports that received a reply.

To trace influence, we reconstructed each company's reporting hierarchy by recursive traversal — following the boss field from each manager up through their own manager, that manager's manager, and so on, up to four levels. We then asked whether a manager's reply behavior is predicted by the reply behavior of the people above them at each level.

Two models carry the headline numbers. A logistic regression estimates the odds ratio for being a replier given the boss's behavior. An OLS regression with company and team-size controls measures how much of a manager's reply rate the upstream chain explains. Minimum sample sizes were enforced throughout: n ≥ 30 for any statistical claim, and at least 3 companies per segment.

Finding 1 — The influence multiplier

The single strongest predictor of whether a manager replies to feedback is whether their own manager does.

The logistic model puts the odds ratio at OR = 2.37 (95% CI: 1.55–3.62, p = 7.2 × 10⁻⁵). The effect survives across model specifications and remains significant after the company-culture control described below. Something real is happening: the behavior of the person directly above you predicts your own.

Finding 2 — Influence decays with distance

If managers learn from their bosses, does the effect reach further up the chart? Does your skip-level, or your skip-level's boss, shape how you manage? We tested each level separately by adding L2, L3 and L4 reply behavior to the model.

The direct manager dominates. Once you account for the L1 boss, levels above add no significant predictive power. This is not bad news — it is clarifying. The cascade works through direct transmission: your boss models behavior for you, you model it for your reports, they model it for theirs. Every link matters, but the influence is local.

Finding 3 — Two cultures inside one company

Influence is local, but it compounds. When good behavior chains together level after level, an entire branch of the org chart inherits it. When a chain breaks, the branch below the break is cut off. The result is branch divergence — different parts of the same company develop distinctly different cultures.

We grouped managers by their top-level leader, the highest person in their reporting chain, and compared reply rates across those branches:

That is not a typo. Some branches reply to nearly everything. Others reply to nothing. These are managers in similar roles, at similar companies, on the same platform. The correlation between a top leader's personal reply rate and their whole branch's average is r = 0.44.

Why standardizing manager behavior is so hard

The local nature of influence explains a problem HR leaders have wrestled with for decades: why is it so hard to make manager behavior consistent across an organization? The answer is that there is no broadcast mechanism. A CEO cannot directly influence frontline managers. Neither can a Chief People Officer. The signal has to pass through every intermediate layer, and if any link breaks, transmission stops there.

One non-replying manager in the middle of the chart insulates everyone below them. Their reports never see feedback replies modeled, so they do not adopt the practice, and neither do the people they manage. A single broken link can cut off dozens or hundreds of people downstream. The 0% to 88% branch spread is the predictable result: high-performing branches have unbroken chains of good behavior; low-performing branches have a break somewhere.

Why the cascade happens

Two mechanisms explain the pattern.

Role modeling. Managers learn by example. When a boss responds thoughtfully to feedback, the manager below internalizes that as "what managers do here." When the boss ignores it, that lesson lands too. The learning is rarely conscious. It is absorbed through hundreds of small observations — the reply notification, the follow-up conversation, the sense that feedback matters.

Cultural reinforcement. Beyond individual modeling, branches develop norms. When everyone around you replies, not replying feels conspicuously wrong. When no one does, replying feels like wasted effort. These norms compound: new managers joining a high-reply branch adopt the behavior within weeks, while those joining a silent branch never start.

What this means for HR

The cascade has a clear strategic consequence: training senior leaders has multiplied returns, because each leader you reach becomes a model for everyone directly below them. But the local nature of influence means top-down training alone is not enough. Three decisions should use this finding directly.

The simplest version of the rule: you do not need to train the CEO, but you do need a good role model directly above every manager you care about. Culture flows downhill. Make sure what is flowing is worth spreading.

Limitations

This study measures association, not proven causation. A few caveats shape how far the findings travel.

• Correlation, not causation. We cannot prove managers learn from their bosses. An alternative explanation: organizations that promote responsive managers build chains of responsive managers through selection. The cascade might partly be a filter.
• Company effects. Managers in high-reply companies reply more regardless of their boss. The model includes a company baseline control, and the upstream chain still explains 32.5% of variance — but some unmeasured company-level factors may remain.
• Survivor bias. We only observe managers who stayed. If unresponsive managers leave high-reply branches, or vice versa, the cascade could partly reflect retention patterns rather than learning.
• Single behavior. The reply rate is one observable manager behavior. Whether the cascade generalizes to other management practices is a reasonable hypothesis, not yet a tested result.
• Smaller upper-level samples. The L3 and L4 estimates rest on a smaller sub-sample of managers with chains that deep, so the absence of effect above L1 is best read as "no signal," not a precise zero.