Why the Pattern Stays
The U.S. studies do not stop at the founding year. They ask what happens to founder–worker composition as a new firm adds people, retains some early employees, and either grows or exits. This chapter follows that lifecycle. Its practical value is a sharper interpretation of persistence: a workforce can remain connected to its founders through retention, later hiring, or continuing organizational influence, and those channels should not be treated as the same thing.
From founding imprint to long-run composition
The balanced LEHD sample follows firms from year zero through year five. Because every firm in that panel has non-founder employees in each of those years, it is well suited to asking how the composition of a surviving, continuously observed workforce changes. The cost is selection. A venture that closes, stops employing people, or fails to meet the data requirements is not part of that balanced trajectory. The estimates describe firms that remain visible and operating, not the average path of every new business.
To look farther ahead, the researchers use unbalanced panels. These include firms even when they are not observed in every later year and therefore allow the analysis to extend to year nine in LEHD and to twenty years in the LinkedIn–Crunchbase data. This solves one problem and creates another. The longer horizon includes more of the firms that disappear, but a decline in the association can now reflect two different events: a surviving firm may become less compositionally tied to its founders, or firms with particular founder backgrounds may leave the sample at different rates. “Attenuation” in the unbalanced panel is therefore not a pure measure of organizational convergence.
Even with that qualification, the association remains substantial. In the unbalanced LEHD analysis, the female coefficient is approximately 0.381 in the founding year and 0.250 in year nine. The corresponding Black coefficient moves from about 0.566 to 0.433. In the longer LinkedIn–Crunchbase panel, the year-zero estimates are about 0.149 for female, 0.085 for Black, and 0.216 for Hispanic workers; by year twenty they remain about 0.135, 0.062, and 0.174. The LinkedIn values are not directly comparable in magnitude to the LEHD values because the samples and demographic measures differ. Their importance is temporal: the association is still visible long after the initial hiring window.
Persistence does not mean that nothing changes. The baseline shares of female and minority workers rise in new firms as they age, and the founder–worker coefficients can decline. A firm may become more diverse overall while still having a workforce composition that is more closely related to its founders than a comparable firm does. The relevant question is not whether the founder imprint remains unchanged. It is whether it remains detectable after the organization has had time to recruit, replace, and formalize its workforce.
The coefficient is a useful way to express that remaining relationship, but it is not a headcount. A female coefficient of 0.250 in year nine means that differences in founder share are still associated with differences in employee share at that age; it does not mean that one quarter of the workforce is female or that one quarter of later hires came through a founder preference. Likewise, the movement from 0.381 to 0.250 is evidence of attenuation in the estimated association, not a direct count of workers who changed jobs. The same discipline applies to the LinkedIn–Crunchbase year-twenty estimates. They show that the slope remains visible in that selected population, not that a particular startup retains the same employees for two decades.
There are two reasons a slope can remain visible. The first is cumulative composition: early hires are still part of the workforce, so the organization carries forward some of its initial conditions. The second is institutional influence: founders may continue to affect key appointments, role definitions, or the processes through which later candidates are assessed. Those explanations make different predictions about what should happen when early employees leave, when a founder steps back, or when the firm introduces more formal recruiting. The available evidence narrows the possibilities, but does not observe all of those transitions directly.
The lifecycle has two kinds of movement. Employees can move through the organization—some leave, some stay, and new people arrive. Firms can also move in and out of the observed data. A long-run coefficient incorporates both. That is why the balanced and unbalanced analyses should be read together rather than ranked as if one were simply a more accurate version of the other.
Retention, initial employees, and the founder’s continuing role
The clearest evidence about the source of persistence comes from comparing the whole workforce with new hires. In both the LEHD and LinkedIn–Crunchbase analyses, homophily among new hires declines more sharply over time than homophily among all employees. The workforce stock therefore remains more strongly associated with founder composition than the later hiring flow does.
Consider a simple example. Suppose a founder’s initial team contains a relatively high share of workers from the founder’s background. If those employees stay, the firm’s total workforce will continue to show that composition even if subsequent recruitment becomes more mixed. The later hiring flow may look less similar to the founder while the accumulated workforce still looks more similar. A persistent workforce association is thus partly a memory of earlier employment decisions.
This stock-flow distinction changes how “the founder keeps hiring similar people” should be interpreted. The data do not support that statement as a complete explanation. They show that the composition of all employees remains connected to founders more strongly than the composition of new hires alone. Retention is not a passive detail; it is one way an early organizational choice can continue to affect access to work years later.
Imagine a firm that begins with ten employees and then adds ten more over the next few years. If six of the original employees remain, the final workforce can still look strongly like the founding organization even if the later ten are more mixed. If most original employees leave, the same later hiring stream will matter much more. The example is simple, but it exposes why a workforce share is a stock built from earlier flows. A persistent association can therefore reveal a durable organizational imprint without implying that every new applicant encountered the same hiring screen.
The distinction also affects how a manager or researcher should read a later workforce snapshot. A diverse-looking set of recent hires may coexist with a less diverse senior or total workforce if earlier employees remain concentrated in influential roles. Conversely, a founder-linked composition among all workers may coexist with a more open current recruiting process. The studies do not provide a complete internal personnel history for every firm, so they cannot convert these possibilities into firm-level diagnoses. They do establish why the flow and stock should be measured separately.
The researchers also compare founder characteristics with the demographic composition of the initial non-founder employees in a horserace regression. During the early years, initial employee composition strongly predicts later employee composition. That is unsurprising: some of those people remain, and early employees may participate in subsequent recruiting. As the firm ages, the predictive force of the initial employee mix fades, while the founder coefficient strengthens significantly. The two effects are not mutually exclusive. Early employees help carry the organization through its first transitions, while the founder’s influence becomes more visible relative to that initial mix later on.
Several mechanisms could produce this pattern. Founders may retain decision rights over key appointments. They may establish routines or standards that affect who fits later roles. They may create production processes in which certain workers are more effective together. Or firms with different founder backgrounds may select into different organizational paths that are not fully captured by industry and location controls. The longitudinal evidence tells us that the relationship survives transitions; it does not identify which mechanism dominates.
The pattern also reaches roles that matter for a venture’s capability, not only the easiest jobs to fill. In LEHD, the association is large across all within-firm pay quintiles and is generally stronger among the highest-paid workers. In the LinkedIn data, the estimates are similar for senior and junior workers. Black and Hispanic workers make up smaller shares of the senior pool, so a similar coefficient can represent a larger relative difference from a lower baseline. The point is not that founders apply one uniform rule to every job. It is that founder-linked composition is visible in different parts of the internal hierarchy.
The occupation results reinforce that interpretation. Positive associations appear among administrators, engineers, finance workers, marketers, operations staff, sales workers, and scientists. The relative effects are particularly large for engineers, a consequential result in a sample tilted toward technology startups because technical staffing can determine what a young firm is able to build. The occupation evidence comes from LinkedIn-derived roles and therefore carries that dataset’s selection limits, but it makes a low-level-only explanation harder to sustain.
Firm characteristics do not erase the relationship either. Service and non-service firms show similar patterns. The fastest-growing ten percent of LEHD firms have nearly identical homophily estimates to the remaining ninety percent. Broad software, hardware, manufacturing, and health-care groups show the same qualitative direction in the LinkedIn–Crunchbase sample, although smaller industry cells are less precise. Estimates are also broadly stable across founding cohorts. This consistency matters because it suggests the pattern is not simply a temporary feature of one startup wave or one narrow business model.
At the same time, consistency should not be mistaken for universality. The estimates are not identical across industries, and a small cell can make a point estimate unstable. The LinkedIn occupation and seniority measures are available only for people with profiles. The fastest-growing-firm comparison is based on realized employment growth, which is an outcome that may itself reflect founder composition, resources, or survival. These subgroup results establish reach, not a single mechanism.
The location of the association changes its significance. If it appeared only among the least-paid workers, one plausible interpretation would be that founders rely on informal or low-cost networks for routine tasks. Its presence among the highest-paid workers and across engineering, science, finance, and operations suggests that the founder-linked pattern can reach roles that determine a venture’s technical capability, managerial capacity, and direction. That does not tell us whether similarity improves performance or limits a firm’s access to talent. It tells us that the distribution of consequential roles is part of the question.
The growth comparison adds a related boundary. The fastest-growing firms show nearly the same qualitative association as the rest, so the pattern is not confined to ventures that never move beyond a tiny founding team. But realized growth is not an external test of founder influence. A founder’s background may affect growth, and growth may affect the ability to formalize recruiting or retain workers. The comparison is therefore descriptive: it shows that rapid employment expansion does not automatically erase the association. It does not show that similarity helps or harms growth.
The cohort results make a different contribution. Estimates are broadly stable across the U.S. founding cohorts studied, with some demographic coefficients moving modestly up and others down. There is no clear evidence in these sources that later startup generations have simply outgrown the pattern. Nor is there evidence that the pattern is fixed at one historical level. Stability across cohorts is best understood as a reason to investigate institutional and organizational mechanisms rather than assume a short-lived fashion.
Path dependence without inevitability
The most defensible interpretation is organizational path dependence. Founders influence the starting conditions of a new firm; initial workers can reinforce those conditions; retention preserves part of the accumulated workforce; and the founder remains relevant as the organization evolves. A small initial difference can therefore have a larger labor-market footprint than the founding team alone would suggest.
But path dependence is not destiny. The coefficients decline in some settings, employee shares change, and new-hire homophily weakens more than overall workforce homophily. The firm can become more diverse without becoming compositionally independent of its founders. Nor does persistence prove a fixed founder preference, a coherent culture, or intentional exclusion. Those interpretations require evidence about decisions, candidate pools, referrals, productivity, and employee movement that these analyses do not directly observe.
This is why the long-run result is an important transition in the argument rather than a final explanation. It tells us that founder background remains relevant after the founding moment, including in higher-paid and technical roles. It also tells us what the next question must be. If similar-background workers remain concentrated in firms associated with similar-background founders, does that reflect workers sorting toward those firms, firms demanding those workers, or an organizational production process that makes the match valuable? The wage evidence is designed to move from persistence to that mechanism question.