What the Evidence Says an Ecosystem Needs
The evidence about Spain’s startup ecosystem becomes useful only after we decide what “entrepreneurial activity” means. This discussion examines a large regional meta-analysis to answer two related questions: which ecosystem conditions are consistently associated with business formation, and which additional conditions are associated with productive, innovative or high-growth entrepreneurship? The practical value is a way to read Spain’s finance gap without mistaking a correlation for a policy guarantee or a broad startup count for a scaleup pipeline.
The measurement problem comes first
Entrepreneurial ecosystems are often described as if they were machines with a fixed set of parts. Add enough finance, talent, networks, infrastructure, culture and support, the argument goes, and entrepreneurial activity should appear. That language is attractive because it makes a complicated social and economic environment sound manageable. It also hides two unresolved questions: which parts matter most, and what kind of activity are they supposed to produce?
The meta-analysis by Martin Queissner, Lennard Stolz and Matthias Weiss was designed to make those questions more empirical. The authors searched academic databases using broad combinations of terms related to entrepreneurs, ecosystems, new firms, entry and activity. They screened 9,435 studies, added forward and backward citation searches, and finally assembled 257 regional-level studies. The resulting dataset contains 431 entrepreneurial-activity data points and 2,241,813 space–time observations. That scale does not make the analysis infallible, but it is a substantial improvement over choosing a handful of famous ecosystems and treating their visible features as universal requirements.
The regional boundary matters. The authors included variables measured at a lower administrative level—such as regions or comparable territorial units—and excluded studies operating only at the country or federal level. This keeps the unit of analysis closer to the places where founders meet workers, customers, universities, investors and support organisations. It also means that the results should not be read as a direct estimate for Spain as a whole. A regional association can identify recurring conditions; it cannot tell us whether a national reform will have the same effect everywhere, or whether a successful Madrid intervention will transfer unchanged to a peripheral Spanish region.
The outcome measures are equally important. General entrepreneurial activity includes new-firm formation, self-employment and total early-stage activity. Productive entrepreneurial activity is narrower: it includes measures such as innovative, knowledge-intensive or high-growth firms. The distinction is not cosmetic. A person may open a business because self-employment is the best available option, while another founder builds a technology company intended to employ hundreds of people. Both are entrepreneurs, but their financing, skills, markets and institutional needs differ.
The authors converted the independent variables from the underlying studies into ten ecosystem categories. Two authors coded the variables separately, discussed disagreements until consensus, and reported 82% intertemporal reliability. They then estimated average relationships using random-effects models, which allow the underlying association to vary across studies rather than assuming that every region is governed by one identical process. The reported values are estimated correlations. A value of 0.21 does not mean that a 10% increase in finance creates a 2.1% increase in startups. It means that, across the collected evidence, the two measures tended to move together with a particular average strength.
That distinction should guide the rest of the argument. The meta-analysis is strong evidence against casual intuition—against assuming that every attractive ecosystem feature is equally important—but it is not a shopping list of guaranteed interventions. Its estimates are also bounded by the variables that earlier researchers happened to measure. If “network quality” is represented by firm density, a null result may tell us more about the proxy than about the value of a trusted introduction between a founder and a customer.
General activity and productive activity are different outputs
The broadest result is that an ecosystem’s conditions cannot be judged independently of its desired output. The total sample shows positive average relationships for several elements, but the patterns change when the data are separated into general and productive activity.
Consider finance, talent and demand. For the total sample, the estimated correlations are about 0.21 for finance, 0.16 for talent and 0.23 for demand. In the general-activity subset, they are approximately 0.17, 0.16 and 0.21. In the productive-activity subset, finance rises to about 0.33, demand to 0.28, while talent remains about 0.16. Each estimate is positive in both outcome groups. The authors therefore identify these three as generic components: market opportunity, financial or risk capital, and human capital form a recurring base for entrepreneurial activity of different kinds.
But a generic base is not a complete theory of growth. It tells us what makes it easier for ventures to emerge; it does not tell us what helps a venture become innovative, knowledge-intensive or high-growth. On that question, the pattern changes. Knowledge has an estimated relationship of about 0.17 with productive activity but roughly −0.01 with general activity. Support services and intermediaries show an estimated relationship of about 0.39 with productive activity but only about 0.02 with general activity. Culture is about 0.33 for productive activity and about 0.11 for general activity, with the latter estimate only marginally separated from zero in the main model.
These figures should not be used to announce that support services are nearly twice as important as finance in a mechanical sense. The productive subset is smaller, the measures are heterogeneous and the confidence intervals are wide—for example, the productive support-services estimate ranges from about 0.13 to 0.60. The more defensible interpretation is structural: knowledge, culture and support appear more relevant when the outcome requires more than opening a firm. They are part of the conversion layer between entrepreneurial potential and productive performance.
Knowledge can provide technologies and opportunities, but only if firms can absorb and apply it. Culture can make entrepreneurship a credible career and make failure survivable, but it also influences whether universities, corporations and investors are willing to engage with uncertain ventures. Support organisations can reduce entry barriers and provide advice, but their value depends on whether the advice is appropriate to the venture’s stage and sector. In each case, the element matters through a relationship, not merely through its presence.
The same distinction changes how formal institutions should be read. In the total sample, formal institutions have an estimated correlation of about 0.36 with entrepreneurial activity, and in the general subset about 0.35. In the smaller productive subset, the estimate is about 0.42 but its confidence interval includes zero. This does not show that rules cease to matter for innovative firms. It shows that the available productive-activity evidence is less precise and that broad measures such as governance, taxes and rule of law may affect the overall economic landscape more directly than they capture the specific mechanisms of a deeptech spin-off or high-growth company.
The lesson is simple but consequential: an ecosystem can be good at generating entry and still be weak at generating productive scale. If a policy is evaluated only by new registrations or early-stage activity, it may look successful while leaving the knowledge, customer, talent and management conditions of growth unresolved.
The evidence can be represented as a map of outcomes rather than as a league table of ecosystem elements:
This map does not claim that the arrows are equally strong or that they describe a proven causal sequence. It makes two boundaries visible. First, the generic conditions recur across outcome types. Second, the conditions associated with productive activity are not simply the same list with more funding added. The empirical question is what combination is missing at a particular place and stage.
Three persistent conditions are complements, not a ranking
It is tempting to turn the meta-analysis into a podium: demand first, finance second, talent third. That would miss the point. Demand, finance and talent are better understood as mutually constraining conditions.
Finance without demand can fund a technically impressive product that no buyer urgently needs. Demand without talent can produce a market opportunity that no team can serve at the required quality or speed. Talent without finance can leave capable founders unable to build, hire or survive long enough to learn. These are not three independent switches. They interact over time, and a shortage in one can change the value of the others.
This is particularly visible in the difference between a general venture and a productive one. A local service firm may begin operating with modest capital, skills available in its immediate labour market and an identifiable local customer base. A research-intensive company may need years of experimentation, specialised staff, regulatory approval, patient capital and a buyer willing to test an unproven technology. The same word—startup—covers both, but an average finance measure cannot capture the difference in their production process.
The authors’ proposal is therefore not that every ecosystem must be designed in one identical way. Their first proposition is that any type of entrepreneurial activity needs generic components such as demand, finance and talent. Their second is that an ecosystem seeking productive entrepreneurship needs additional knowledge, culture and support components. Their third proposition adds a contextual boundary: the form and quantity of the generic components depend on the region, and the combination of elements influences the ecosystem’s primary output.
That final point prevents a common policy error. A country cannot simply import the visible institution of a successful ecosystem and expect the same output. A large fund may be appropriate where a region has a deep technical labour pool, strong customers and a route to international markets. In another region, the binding constraint may be the absence of those customers or the inability of universities and firms to transfer knowledge. The fund is not irrelevant in the second case; it is simply unlikely to be the first missing link.
What a weak estimate does—and does not—tell us
Networks and physical infrastructure are the clearest test of why evidence must be interpreted rather than recited. In the main total-sample results, networks have an estimated correlation of −0.04 and physical infrastructure 0.08; neither is statistically distinguishable from zero. The same broad non-significance appears across the general and productive subsets.
A superficial reading would say that networks and infrastructure do not matter. The authors do not make that claim. Network measures in the underlying literature include variables such as firm density, social capital and network ties. Those may not measure whether a founder can reach the particular investor, researcher, supplier or customer who can unlock the next step. A count of nearby firms can be high while the useful connections among them remain weak. An aggregated social-capital measure may capture regional culture rather than the connecting function of networks.
Infrastructure has a different problem. Basic transport, electricity and communications can be essential when they are missing. But once a region reaches a sufficient level, additional roads or capacity may have diminishing effects on business formation. Information and communications infrastructure may still be below the threshold required for a new generation of digital businesses. A null average across regions at different stages of development can therefore conceal a threshold relationship.
The evidence also offers a reason to preserve the system view. Studies that include more ecosystem elements tend to report higher explanatory power for entrepreneurial activity. That pattern is consistent with the idea that several endowments and their interactions explain more than any one variable alone. It does not prove that the elements cause one another, or that a larger model is automatically better. It does show why a single-input diagnosis is too narrow.
Leadership illustrates the opposite problem: not too much measurement, but too little. The total estimate is about 0.34 and the general-activity estimate about 0.32, both moderate positive associations, yet they come from only 17 and 14 studies respectively. There are not enough productive-activity observations to make a general claim. Leadership may be connected to culture, talent and networks, and visible founders may matter as role models or mentors. The evidence base is simply not strong enough to give leadership the same status as the more extensively studied generic conditions.
The distinction between absence of evidence and evidence of absence is not a ritual caveat. It changes what a decision maker should do next. A weak network estimate should prompt better measurement of introductions, collaborations, pilots and resource movement. A threshold-sensitive infrastructure estimate should prompt identification of underserved places or technologies rather than indiscriminate expansion. A thin leadership literature should prompt better data collection before a confident national programme is justified.
The evidence changes the finance question
Taken together, the meta-analysis does not say that finance is overrated. It says finance is one of the conditions that repeatedly travels with entrepreneurial activity, and it appears particularly relevant to productive activity. That is precisely why the capital-only argument is too weak. If finance were irrelevant, the diagnosis would be easy: stop worrying about it. If finance were sufficient, the remedy would also be easy: add money. The evidence supports neither conclusion.
It supports a conditional claim. Capital has more chance of producing productive activity when it meets demand, talent and the additional conversion conditions of knowledge, culture and support. The conditions need not be equally strong everywhere, and their effects may vary by stage and outcome. The empirical literature cannot determine the exact combination Spain should choose, but it can discipline the question Spain asks of its own ecosystem: is the problem a missing resource, a weak connection between resources, a mismatch between stage and instrument, or a definition of success that counts activity rather than productive growth?
That is the standard against which Spain’s finance data should be read. The country-specific diagnosis can now move from a general evidence hierarchy to a concrete lifecycle: credit and seed capital at the beginning, the larger rounds required for scale, and the investors and institutions that must connect those stages. Only then can we tell whether Spain needs more capital in general—or a better functioning route from early promise to durable scale.