A control group study of incubators’ impact to promote firm survival

Abstract

It is widely unclear as to whether start-up firms supported by publicly-initiated incubator initiatives have higher survival rates than comparable start-up firms that have not received support by such initiatives. This paper contributes to the underlying discussion by performing a large-scale matched-pairs analysis of the long-term survival of 371 incubator firms (after their graduation) from five German incubators and a control group of 371 comparable non-incubated firms. The analysis covers a 10-year time span. To account for the problem of selection bias, a non-parametric matching approach is applied to identify an appropriate control group. For neither of the five incubator locations, we find statistically significant higher survival probabilities for firms located in incubators compared to firms located outside those incubator organizations. For three incubator locations the analysis reveals statistically significant lower chances of survival for those start-ups receiving support by an incubator. The empirical results, therefore, raise some doubts regarding the impacts of incubation on long-term firm survival.

1 Introduction

For more than half a century, publicly funded incubation initiatives are at the heart of urban and regional technology and innovation policies for the promotion of entrepreneurship, to support first-stage development of new technology-based firms, to strengthen academic–industry linkages, and to promote innovation activities (European Commission 2002; Hackett and Dilts 2004; OECD 1997). Federal, state and local funding agencies worldwide have been, and still are, very enthusiastic in establishing incubation facilities, such as business incubators, science parks or technology centers. And in most cases they do this without profound knowledge of their actual effectiveness. Incubation, and the long-term outcomes of incubation can still be considered a ‘black box’ (Hackett and Dilts 2008).

However, what clearly can be observed is a major change in awareness by both academics and policy actors in favour of rigorous empirical evaluations to demonstrate long-term incubation industry’s effectiveness. It is increasingly acknowledged that tracking the development of incubated firms beyond the initial incubation phase, that is the post-graduation period, is critical for the understanding of overall incubators’ effectiveness. This change in awareness is, most notably, reflected in a recent initiative that is undertaken by the National Business Incubation Association (NBIA 2007). This initiative seeks to encourage managers of US incubation programs to systematically and periodically collect outcome data of incubator graduates to track the long-term performance after they have left the protective environment of their incubating organization.¹ Of prime interest within this program is the collection of data on graduate survival, since those statistics are considered to be most significant for providing evidence on incubators contribution.² That said, NBIA (2007) concedes: ‘The fact is, no one has ever compared the survival of incubated versus non-incubated firms.’ Indeed, the limited evidence to date provides little insights regarding the contribution of incubation towards long-term firm survival.

Given that the promotion of long-term survivability is one of the key objectives of incubation, the lack of systematic research on incubator-firm survival is astonishing. This disregard is strongly dependent on the challenges associated with survival rates as variables of incubators’ effectiveness (Phan et al. 2005; Schwartz 2009; Sherman and Chappell 1998—these will be discussed in Sect. 2.3). There also exist deficits concerning systematically recorded data on formerly incubated firms (Colombo and Delmastro 2002; Hannon and Chaplin 2003). The crucial question, as to whether incubation acts as means to overcome the resource deficiencies start-up firms face in the first years after the market entry (Hannan and Freeman 1984; Stinchcomb 1965), can only be answered by tracking the performance of incubated ventures after they have completed their incubation period (Hackett and Dilts 2004; Hannon and Chaplin 2003; Rothaermel and Thursby 2005; Schwartz 2009). Although, specifically in Germany, scholars are cautiously optimistic regarding the survival benefits of incubation (Elle et al. 1997; Pleschak and Werner 1999; Seeger 1997), an empirical analysis is currently non-existent.

The present paper responds to the aforementioned deficits by conducting an analysis of the long-term survival of 371 incubator firms (after their graduation) and contrasting these results with the long-term survival of a control group of 371 comparable non-incubated firms. Previous studies on incubator firm survival either cover comparably short time spans or have relatively small sample sizes. Whereas virtually all existing studies on incubator firm performance use cross-sectional data, the present study covers all firms that have graduated since the inception and track them over time, thus avoiding sampling bias (Siegel et al. 2003). We contribute to the literature in several important ways. First, most attempts to compare incubator firm survival with the survival of non-incubated firms fail to identify appropriate control observations retrospectively, because official data sources do not maintain information on firms that have been subject to a market exit (Sherman and Chappell 1998). The key advantage of our data (Creditreform) for this study is that, once registered, firms do remain in this database even if they have been subject to a market exit. Second, since incubation is not random, a non-parametric matching approach is applied to identify an appropriate control group of non-incubated firms. Third, the unique data set allows us to apply techniques of survival analysis to broaden the analysis from a pure static comparison of survival/closure rates towards a dynamic exploration of market exits. We also take account of right censoring as well as left truncation.

The following section presents our analytical framework, including a brief theoretical debate on the causes of high failure risks of start-ups and how incubators are expected to contribute to its reduction (Sect. 2.1). Moreover, we overview the incubation industry in Germany from 1983 to 2009 (Sect. 2.2) and discuss the main methodological problems associated with survival rates (Sect. 2.3). Section 3 reviews empirical results regarding survival of incubator firms. Data and methodology are described in Sect. 4. Empirical results of incubator firm survival versus survival of non-incubated firms are presented in Sect. 5. Discussion is held in Sect. 6.

2 Analytical framework

2.1 Why start-ups fail—and how incubators are expected to reduce the risk-of-death

Incubator organizations (business incubators, science parks, research parks, etc.) theoretically can be understood as policy-driven instruments to respond to the ‘liability of newness’. Organizational ecologist use the notion of a ‘liability of newness’ to describe the considerable high failure risk that young or newly founded ventures face in the very first years after their market entry, because they do not (yet) possess enough or the necessary resource portfolio they need to survive (Brüderl and Schüssler 1990; Freeman et al. 1983; Hannan and Freeman 1984; Stinchcomb 1965). According to this hypothesis between age-dependency and firms’ death risks, start-up exit dynamics are characterized by a hazard rate that is highest immediately after market entry but decreases monotonically over time. Similar arguments are made by supporters of a ‘liability of smallness’—hypothesis. Instead of focusing on age-dependency and firms’ death risks, this concept focuses on the impact of organizational size on firms’ death risks (Brüderl and Schüssler 1990), as an alternative explanation for declining failure risk with time (Geroski 1995).

The basic argument for both hypotheses is that (small) start-up firms are confronted with a considerable discrepancy between key resources that are crucial for their viability and their actual resource base. Larger, or more mature, firms can draw on a broader pool of resources to overcome problematic business situations. The amount of initial resources and endowments increases with firm size (Aldrich and Auster 1986; Brüderl and Schüssler 1990). Stinchcomb (1965) points out that new organizations lack stable business relationships (with customers or suppliers), must learn their specific roles and tasks as social actors and have to develop routines—a process which takes time and involves considerable costs. Furthermore, start-ups and young firms in particular do not possess any reputation and need some time to gain legitimacy in the market. At worst, they are associated with a rather negative image due to their novelty or because they have new products/services. According to Hannan and Freeman (1984), firms with low reliability and accountability will be eliminated from the population. Therefore, young firms need to demonstrate that they are reliable and trustworthy business partners towards their environment (including e.g. customers, suppliers, or investors).³ Subsequently, we illustrate why and how incubation should contribute to the reduction of high start-up mortality.

The basic tangible and intangible ingredients of incubation (summarized in Fig. 1) have been listed and discussed in numerous studies (e.g. Allen and McCluskey 1990; European Commission 2002; OECD 1997; Sternberg et al. 1997; Westhead and Storey 1994). Spatially concentrated, subsidized rental space, including offices, laboratories, and small production space, helps to reduce early-stage fixed costs through co-location and shared utilisation (Aerts et al. 2007; Chan and Lau 2005; McAdam and McAdam 2008). Flexible leasing terms ensure the possibility to extend/reduce the space occupied by the firms immediately, depending on their actual business situation. A broad spectrum of collectively shared facilities and services (secretarial support, laboratory equipment, communication infrastructure, etc.) constitute the second element. Particularly small firms or one-man businesses as well as firms from R&D-intensive sectors, such as biotechnology, face severe difficulties if they do not have access to such essential but often extremely expensive facilities and sophisticated equipment.

Fig. 1

The basic elements of incubation

Business assistance services (for example, marketing, accounting, human resources) constitute another cornerstone of incubation (Rice 2002), in order to promote the understanding of vital day-to-day business processes. These services assist the incubated firms in areas where they do not possess the relevant knowledge and expertise (e.g. Allen and McCluskey 1990; European Commission 2002; McAdam and McAdam 2008). Managing resources and expertise, including business experiences, are key factors in identifying, combining and exploiting the economic potential of the resource endowment of the firm (Barney 1991; Mahoney 1995). A firm’s image, and its reputation, (Fischer and Reuber 2007), are highly valuable intangible resources (Barney 1991). Naturally, start-ups and young firms in particular do not possess any kind of reputation or legitimacy in the market. This might have a negative effect on a variety of business interactions—for example, negotiations with suppliers, customers or financing institutions. Obtaining the benefits of an image associated with an incubator location and acquiring credibility is another important mechanism of incubation (Ferguson and Olofsson 2004; McAdam and Marlow 2007; McAdam and McAdam 2008).

Efficient networking within the incubators that fosters co-operative, formal contract agreements as well as informal interactions between the incubatees is seen as decisive for a successful incubation process (Bøllingtoft and Ulhøi 2005; McAdam and McAdam 2008). As social capital theory states, social networks among entrepreneurs are seen as a critical strategic resource (Granovetter 1985). Spatial proximity between incubator firms facilitates the transfer of information, knowledge and the exchange of experiences. Additionally, the incubator takes the position of an intermediary (Scillitoe and Chakrabarti 2010), helping start-ups to establish contacts to incubator-external actors and to gain access to their resources and knowledge. Incubators’ network may comprise potential customers and suppliers, a wide network of specialized service providers (e.g. lawyers, tax accountants), financial institutions (e.g. banks, venture capitalists), public and private research facilities and political institutions (e.g. local development agencies, funding agencies). Such cooperative relationships can themselves even be understood as critical intangible resources (Gulati 2007; Uzzi 1997). Specifically linkages to academic institutions are relevant for technology-based firms to keep abreast of, and to acquire the most recent scientific knowledge (Link et al. 2007).

Through the interaction of these support mechanisms, incubators focus on the compensation of fundamental early-stage resource deficits of start-ups to ensure entrepreneurial stability, sustainable economic growth and long-term business survival. It is therefore expected that incubator support may increase long-term survival chances of incubated ventures.

2.2 Incubation in Germany—a brief overview

There are about 400 incubator organizations in operation at the end of 2009 in Germany (Schwartz and Hornych 2010). Since 1983 when the first incubator opened in Berlin, high commitment by cities and municipalities in particular has led to the largest and one of the densest populations of incubators in Europe (European Commission 2002). As the result of this high regional distribution, the number of newly opened incubators is decreasing since the end of 1990s. Similar patterns are reported for European incubators in general (Aerts et al. 2007) reflecting unfavorable economic conditions at the beginning of the 21st century (for example, the decline of investments; insufficient start-up activity).

Incubation facilities in Germany are officially termed ‘Innovation Centres’ by the German Association of Technology Centres (ADT), covering two types of incubator organizations: business incubators and technology centres. While business incubators focus mostly on newly-founded ventures with little differentiation regarding technological sophistication, technology centres mainly focus on innovative small and medium-sized firms (Baranowski et al. 2008; Sternberg et al. 1997; Tamásy 2001). One could say that business incubators aim at the reduction of the ‘liability of newness’ and technology centers aim at the reduction of the ‘liability of smallness’. However, this would be an idealized view because both types do not restrict themselves to one target group or another and provide comprehensive support as detailed in Sect. 2.1. Thereby, the name of an incubator organization does not necessarily reflect its ‘content’ (Tamásy 2001).⁴ For instance, 16 % of German incubator facilities call themselves ‘Science/Technology Parks’ (Baranowski et al. 2008), but these facilities must be considered business incubators or technology centres actually. For example, linkages to academic institutions are less pronounced than it is the case for science parks, such as in the UK (e.g., Westhead and Storey 1994), the US (e.g., Link and Scott 2003) or in Sweden (e.g., Löfsten and Lindelöf 2002). Only few German facilities maintain formal linkages with higher education institutions (Sternberg 2004; Tamásy 2001). Overall, the German concept of incubation facilities corresponds closely to the definition of business incubation put forward by the UKBI (2009). To avoid juggling with names, in the present study the term incubator will therefore refer to both German constructs. On average, German BI organizations provide rental space of 5,964 m² in addition to 900 m² of infrastructure space and have 33 tenant firms with seven employees each (Table 1).

Table 1 Incubation in Germany

2.3 Some notes on incubator firm survival rates

This section shows those main challenges associated with survival rates, and failure rates alike, as indicators of incubator effectiveness that usually prevent scholars to tackle issues of incubator firm survival. These—partially interconnected—challenges are summarized in Fig. 2 and are briefly discussed below.

Fig. 2

Challenges associated with incubator firm survival as indicator of the effectiveness of incubators

First, survival rates are primarily criticized for their limited explanatory power, because of the systematic selection processes applied by the incubating organization prior to the incubation (1). Applying firms are screened according to a number of criteria (e.g. business plan, financial budget qualifications of the entrepreneur, properties of the product) to assess their future success chances. The result is a considerable selection-bias (Bearse 1998; Phan et al. 2005; Sherman and Chappell 1998). Firms with a high risk of not surviving do not usually pass the admission criteria. This incubator-specific selection process may induce relatively low failure rates, suggesting an overestimation of the effectiveness of incubators in this regard. Second, this endogeneity simultaneously limits the explanatory power of so-called tenant survival rates (2). In Germany for instance, according to official data from ‘ADT’, the survival rate during incubation is more than 90 %. However, ‘since incubators are specifically designed to maintain and increase life span’ (Phan et al. 2005) it is essential to include the firm after graduation from the supporting incubator when analyzing survival processes (Schwartz 2009).

Third, scholars refrain from using survival rates, because as sole indicators of incubator performance these measures are insufficient to capture the success of a particular incubator (3). They may cover only one dimension of the complex support process (Bergek and Norrman 2008; Hackett and Dilts 2004; Schwartz and Göthner 2009 for related discussions). For instance, there is no clear understanding whether the total number of firm survivals matter more for incubator assessment than the growth trajectories of the biggest successes.⁵ Survival rates alone, as any other indicator, are insufficient to provide a comprehensive picture of incubator performance. This will be discussed in greater detail in Sect. 6.

Fourth, benchmarking criteria or threshold values that indicate what rate of survival can be considered satisfactory are missing (4). Neither the incubator organizations nor local decision makers (local development agencies, politicians) define such criteria. If anything, these are vague verbalized and therefore difficult to control. Therefore, definite assessments of survival/failure rates are difficult. In this context, incubator idiosyncrasies must also be recognized (as fifth main challenge). The business incubation industry is heterogeneous (Hannon and Chaplin 2003) and incubators are idiosyncratic regarding, for example, their client needs and their regional context and underlying objectives (5). Though, superior economic objectives of incubators are widely comparable and therefore measures of success should be quite similar (Ratinho and Henriques 2010; Schwartz and Göthner 2009), the appropriateness of survival rates as success indicator may vary between different locations (Bergek and Norrman 2008).

Notwithstanding these criticisms, by no means firm survival may be ignored as measure of incubator success. Survival is the minimum criterion of entrepreneurial success and is at first the most important attribute of firm development (Tamásy 2005; Woywode 2004). Accordingly, one of the primary objectives of incubators is the promotion of survivability and the positive development of their tenant companies (Lalkaka 1996; McAdam and Marlow 2007; Schwartz 2009).

3 Prior results on incubator firm survival

Below we give an overview over most important empirical studies that focus on incubator firm survival, with a particular focus on those empirical studies applying a control group methodology. Table 2 summarizes existing results of incubator firm survival. It must be noted that there is no uniform definition of firm closure in the literature. Heterogeneous data regarding the incubators studied (type of incubator, strategic objectives, industry focus, location, university linkages, entry criteria etc.) and firms included (tenants versus graduates, technology orientation, industry affiliation, age, etc.) must be considered as well as varying sample sizes. Overall, there is no perfect comparability between the results.

Table 2 Overview over reported closure/survival rates of incubator firms (in descending order according to time span covered by the analysis)

Whereas few attempts have been made to explicitly explore post-graduation survival (Schwartz 2009; Seeger 1997; Willms and Sünner 2004), most studies consider tenant survival rates (Rothaermel and Thursby 2005) or have mixed samples, i.e. actual tenants as well as graduate firms (Ferguson and Olofsson 2004; Storey and Strange 1992; Westhead and Storey 1994). Furthermore, only one study examines exit dynamics. Taking into account the problems of selection-bias and missing benchmarking criteria, it may be less meaningful to evaluate incubators’ effectiveness by calculating survival/closure rates of incubator firms. One approach to cope with these issues is the application of control group methodologies. Given their potential, one might suggest that control group based studies are frequently used to assess value-added contributions of incubators with respect to firm survival. However, this is not the case mostly because of non-availability of data.⁶ Little is known as to whether incubator-supported firms indeed have a higher probability of survival compared to comparable firms located outside such facilities. Westhead and Storey (1994) investigate performance differences between 183 firms located on UK science parks and a control group of 101 non-science park firms. Over a period of 6 years, 38 % of the on-park firms, but only 32 % of the off-park firms, are considered failures (for their definition, see Sect. 4.3). Therefore, the authors question the impact of UK science parks on firm survival, which corresponds to the scepticism expressed for German incubators (see Sect. 1). Contrasting these results, Ferguson and Olofsson (2004) find that Swedish science park firms have higher survival rates than comparable off-park firms. In their study, the authors investigate the survival of 30 new technology based firms located on Swedish science parks and compare the findings with survival rates of 36 comparable off-park firms. After a period of 7 years, 93.3 % of the firms originally located in science parks were still in operation versus only 66.7 % of the off-park firms. Eleven firms graduated successfully from the science parks and none of them failed. Although, there is a growing body of literature regarding the impact of incubation on firm survival, this brief overview demonstrates that to date there is only vague empirical evidence as to whether these instruments can be considered effective in terms of the promotion of firm survival.

4 Data collection and methodology

4.1 The ‘On-incubator’ sample

We initially chose to restrict the present study to five incubator locations. This selection primarily was driven by the age of the particular incubator organization. A minimum operation time of at least ten to 15 years is commonly assumed to be essential to achieve reliable evaluation results (Autio and Kauranen 1992; Schwartz and Göthner 2009). All incubators chosen for this study were established in the early 1990s and exhibit an operation time of at least 13 years at the time this research project was started (spring 2007). Specifically, we include incubators located in the cities of Dresden, Jena, Halle, Neubrandenburg and Rostock.

To collect background information about the five incubators investigated in this study, face-to-face expert interviews were conducted with the manager of each incubator. These personal interviews primarily focused on information with respect to the processes of the establishment (e.g., What were the driving forces behind the establishment of the incubator? What where the regional pre-conditions? Where there any support programs that were used to finance the establishment?). The interview-technique is a commonly used, sometimes additional, method in business incubation research (e.g. Chan and Lau 2005; Monck et al. 1988; Von Zedtwitz and Grimaldi 2006). These interviews were done on the basis of a structured interview guide, but no pre-formulated answers were given in the interviews.

The incubators are particularly comparable in terms of their non-profit orientation, age, maximum tenancy, the market strategy applied and with respect to their main objectives and target groups. All five incubators can be considered managed science-parks with a full-time manager on site (Westhead and Storey 1994). Furthermore, none of the five incubators applies formal exit rules. In some cases there are periodic discussions between firms and incubator management during the course of the incubation process to detect undesirable developments. However, besides maximum tenancy, specific exit criteria (e.g. annual growth/decline in employment) do not exist. A short profile for each incubator is given in Table 3.

Table 3 Main characteristics of the selected incubators

For each of the five incubators, complete lists of all ever incubated and subsequently graduated firms and organizations from incubators’ commencement until December 31, 2006 were submitted by the respective incubator management. A total number of 462 firms and organizations were identified. This database was adjusted by removing 52 non-private organizations (such as university institutions, public research laboratories and local development agencies). Firms that do not rent any physical office space but use the incubator address for their business mail were also excluded. Further 31 firms had to be omitted from the analysis because these firms could not be identified. After these adjustments, 371 firms that have graduated from the five incubators remained for the analytic purpose of the present study. These 371 firms form our ‘On-incubator’ sample.

4.2 The ‘Control group’ sample

The non-parametric propensity score matching (Leuven and Sianesi 2003; Rosenbaum and Rubin 1983; Rosenbaum and Rubin 1985) is applied to identify the ‘Control group’ sample(s). This technique matches incubated firms and non-incubated firms on the basis of important exogenous characteristics, and selects the best statistical twin from a group of potential control observations for each firm from the incubator sample.

Previous studies in incubator/incubation research use four important variables to identify an appropriate control group: firm location, industry affiliation, age of the firm and legal form (e.g., Colombo and Delmastro 2002; Ferguson and Olofsson 2004; Lindelöf and Löfsten 2004; Löfsten and Lindelöf 2002; Westhead and Storey 1994). We adopt this approach for the present study. Data restrictions inhibit the inclusion of further explanatory variables in the matching process (see also the limitations in Sect. 6). There is, however, some evidence for the advantages of relatively few explanatory variables (Caliendo and Kopeinig 2005).

Data (for both samples) were collected using firm-specific information by Creditreform. Creditreform is the largest credit rating agency in Germany and collects detailed information on almost all firms in the German commercial register (see also Almus and Nerlinger 1999). The key advantage of Creditreform data sets for this study is that, once registered, firms do remain in the database even if they have been subject to a market exit. In these cases Creditreform records the exact date of deregistration of business from the commercial register, the date of bankruptcy or the date of the merger contract.

• 1. Firms from the ‘On-incubator’ sample are classified according the two-digit level of the NACE Rev. 2 classification of economic activities. Eight main industry-groups were built (“Appendix 3”).⁷

• 2. Firms were classified into five start-up periods based on their year of establishment.

• 3. The legal form of a firm (at the time the firm was founded) is included as matching criteria. This is important because firms with limited liability might face a greater risk of death due to a higher willingness to pursue risky projects (Almus and Nerlinger 1999).

• 4. Firms’ location is given by the location of the respective incubator, i.e. by cities.

In a next step, Creditreform provided for each of the five locations data on all firms in their database that have been founded up to the end of 2006 and matches one of the main industry-groups defined above. Using the same database for treatment observations and control observations is an important aspect (Caliendo and Kopeinig 2005; Heckman et al. 1999), though often neglected. Over all five locations, 43,467 potential control firms were identified that did not receive public support by an incubator. “Appendix 3” shows that the ‘On-incubator’ sample differs significantly in important characteristics from the group of potential controls. Thus, incubator firms are a selective group of firms and incubator support is not arbitrary.

To avoid biased results due to these significant differences, we apply propensity score matching. For every observation from the ‘On-incubator’ sample and the group of potential controls, the vector of exogenous variables is condensed into one single measure: the so called propensity score. In the present application, the propensity score reflects the likelihood that a firm i has received support by an incubator, conditional on a set of individual characteristics x _i : Pr(S _i  = 1 | X = x _i ), with x _i given by the variables defined above. Propensity scores are estimated using probit models. Relevant exogenous variables (industry affiliation, legal form, start-up period) are regressed on a binary dependent variable indicating as to whether a firm were incubated or not. Location was not included as exogenous variable. Since we are interested in a differentiated analysis, we performed estimation of propensity scores for each incubator location separately. This procedure is particularly recommended if heterogeneous effects for sub-populations are expected (Caliendo and Kopeinig 2005). The respective estimation results are not further discussed here, but are provided in the “Appendices 1 and 2”. According to Caliendo and Kopeinig (2005) all variables are fixed over time, measured at start-up.

There are three general assumptions in the context of matching. First, the conditional independence assumption assumes that all relevant exogenous variables affecting both treatment and outcomes (survival) are observed. Second, the stable unit treatment value assumption demands that treatment of one particular firm (incubation) does not affect the outcome of other firms (survival).⁸ The validity of both assumptions cannot be tested empirically however. Particularly the conditional independence assumption seems problematic, because we might exclude relevant (firm-specific) variables, such as market potential of the underlying business model (see Sect. 6 in detail). The third assumption refers to common support and assures that for every treatment observation, given the relevant exogenous variables, a similar control observation can be identified. To fulfill this assumption, a region of common support (RCS) is defined where propensity score estimates between treatment observations and potential control observations overlap. Observations outside the RCS are discarded.

4.2.1 Sample comparison and quality of the matching result

The left half of Fig. 3 gives the distribution of the propensity scores before the matching procedure for the ‘On-incubator’ sample and for potential controls. The RCS is given by the overlap between the two curves and is plotted via the vertical dashed lines. The RCS was determined by minimum–maximum-comparisons of the distribution of the propensity scores between potential controls and ‘On-incubator’ firms. The RCS varies between the five incubator locations and therefore the requirement to discard observations from the ‘potential-controls’ group. The large number of potential matching partners per incubator location allows for exact matches in most cases (matching without replacement; nearest neighbour). That is control observations are assigned only once to one supported incubator firm, with both observations having the same propensity score. The situation after the matching procedure is depicted by the distribution of the propensity scores for the ‘On-incubator’ sample and the selected ‘Control group’ sample (right half of Fig. 3). Further, “Appendix 3” shows the distribution of important characteristics between both groups. No significant difference for any relevant exogenous variable is observed. This holds for each of the five incubator locations.

Fig. 3

Distribution of the propensity scores before (left column) and after (right column) the matching procedure (Kernel density estimation, Gaussian Kernel); analysis per incubator location. Note: ‘RCS’ region of common support. ‘RCS’ determined by minimum–maximum-comparisons of the distribution of propensity scores between potential controls and ‘On-incubator’ firms. Upper/lower threshold values for the ‘RCS’ are plotted via the dashed vertical lines

4.3 Survival analysis

4.3.1 Definition of firm closure

A firm is identified as closure or survivor at the end of 2006 based upon Creditreform data. This also means that the same data source is used for both samples ensuring homogeneity in the specification of firm closure across ‘On-Incubator’ and ‘Control group’ samples. In line with Westhead and Storey (1994), an independent firm is considered closure, if it is definitely not identifiable as actively trading business at the reference date of December 31, 2006.⁹ Relocations are not regarded as closures, if the firm continues trading at the new location. Changes regarding the legal form are not considered as closures. Subsidiaries (e.g. trading office) are considered closure, if this subsidiary is not identifiable in the incubators’ city (even if the parent company continues trading) or the respective parent company was closed. This definition is chosen to enable direct comparisons with the results of Seeger (1997) or Westhead and Storey (1994) and includes all firms that ended up in liquidation or bankruptcy, firms that have been subject to a merger or acquisition¹⁰ and firms that are not actively trading business in any identifiable form. Considering the latter, Creditreform explicitly marks those firms that are still officially registered but do not actively have any identifiable business activity.

4.3.2 Survivor functions and hazard rates

An important objective of this paper is to shed light on market exit dynamics. We therefore apply statistical methods within a hazard rate framework (Cleves et al. 2004; Kleinbaum 1996; Lawless 1982). Such techniques have the huge advantages of not being restricted to a dichotomous variable of surviving/not surviving, to take into account the precise duration until the market exit and to account for censoring (Ferguson and Olofsson 2004; Schwartz 2009 for applications in the context of incubation research). A firms’ probability of surviving beyond a certain point of time t (measured in years since start-up) is reported by the survivor function S(t) (Eq. 1), with 0 ≤ S(t) ≤ 1 and with T as a non-negative random variable that denotes the time of the event. Survivor functions account for right-censoring, because there may be observations that do not experience the ‘event’ within the observation period. S(t) is estimated by the non-parametric Kaplan–Meier estimator (Kaplan and Meier 1958—[Eq. 2]), where n _j is the number of firms at risk at time t _j and d _j the number of ‘events’ at t _j .

$$ S\left( t \right) = 1 - F\left( t \right) = P\left( {T > t} \right) $$

(1)

$$ \hat{S}\left( t \right) = \prod\limits_{{j|t_{j} \le t}} {\left( {({{n_{j} - d_{j} )} \mathord{\left/ {\vphantom {{n_{j} - d_{j} )} {n_{j} }}} \right. \kern-\nulldelimiterspace} {n_{j} }}} \right)} $$

(2)

For the analysis of the development of closure risk over time, hazard rates h(t) can be considered additionally (Eq. 3). Given our context, the hazard rate is defined as the firms’ probability that a market exit occurs in a given interval [t, t + ∆t] (year after start-up), under the condition of having survived until the beginning of that interval. Hazard rates are also called instantaneous failure rates with 0 ≤ h(t) ≤ 1 (see Cleves et al. 2004; Kleinbaum 1996 for more details).

$$ h\left( t \right) = \mathop {\lim }\limits_{\Updelta t \to 0} \frac{{P\left( {t \le T < t + \Updelta t|T \ge t} \right)}}{\Updelta t} $$

(3)

4.3.3 Left truncation

For the ‘On-incubator’ sample, we must adjust for delayed entry, because we observe incubator firms from their start-up date but they enter our analysis only because they have started incubation. 169 firms from the incubation-sample already survived until the beginning of incubation and, obviously, could not have been failed before. Ignoring this pre-incubation period leads to incorrect values for S(t) and h(t) and a bias towards higher (lower) survival (hazard). We correct for this bias caused by left truncation by omitting firms from calculations in their truncation period (see Cleves et al. 2004; Lawless 1982), leading to a reduction of the total number of observations under risk for each interval considered. Since our dataset contains exact starting dates of individuals’ incubation periods, these firms enter the analysis at the time they start incubation (i.e. when they become at risk).

5 Empirical results

5.1 Survival and closure rates

This first sub-section presents empirical results on static survival rates over the entire observation period. Then, Sect. 5.2 analyzes exit dynamics using a duration analysis framework. Given the underlying definition of firm closure as specified in Sect. 4.3, Table 4 gives an overview over the number of survivors, the number of closures as well as the survival rates and closure rates for each of the incubator samples and the respective control group. We prefer focusing on survival rates since these are independent of different interpretations, i.e. inclusion of M&A-cases as closures or survivors.

Table 4 Static comparison of survivors and market exits between ‘On-incubator’ firms and ‘Control group’ firms (reference date December 31, 2006)

Regarding the ‘On-incubator’ samples first, distinct differences between survival rates can be observed. Whereas the survival rate for firms supported by the TZD amounts to 70.6 %, a comparatively low share of firms being supported by the TIGN of 40.6 % survives long-term irrespective of the protective environment provided by the incubator. Except for the TGZH, a small fraction of 17 formerly incubated firms, accounting for 10.8 % of all market exits, have been the subject of a merger or acquisition. Overall, results in Table 4 underscore the need for incubator-specific analyses.

Considering differences in survival rates between the ‘On-incubator’ samples and ‘Control groups’, a vague distinction between two groups of incubator locations can be made. For incubators located in the cities of Dresden (TZD) and Halle (TGZH), survival rates for both samples are approximately equal. There is also a slight tendency, though not statistically significant, that ‘On-incubator’ firms have higher survival rates than firm not receiving support by incubation. However, the second group of incubator locations, i.e. Jena (TIPJ), Neubrandenburg (TIGN) and Rostock (RIGZ) gives a complete different picture. ‘On-incubator’ survival rates are far below the survival rates for ‘Control groups’ over the observation period. The most striking difference results for the TIPJ, where half of the ‘On-incubator’ firms survive compared to an almost 90 % survival rate for ‘Control group’ firms.

Prior research by Westhead and Storey (1994) claimed that higher closure rates for firms located in UK science parks result from the higher share of subsidiaries in these parks compared to Off-park locations; while survival rates for independent firms inside and outside these parks are virtually identical. Subsidiaries are frequently founded to benefit from relatively low rents, but are often closed by their parent companies with expiration of maximum tenancy. In the present study, the establishment type (subsidiary vs. independent firm) does not make a difference. The aggregated ‘On-incubator’ sample includes 29 subsidiaries with a survival rate for this sub-sample of 58.6 %. The aggregated ‘Control group’ sample includes eleven subsidiaries with a survival rate for this sub-sample of 54.6 %.

A closer look at survival rates according to the eight main industry-groups shows that the biggest gap in survival between ‘On-incubator’ firms and ‘Control groups’ can be observed for ‘Wholesale trade/Retail trade’, ‘Computer’ and ‘Research and Development’. For these three industry groups, survival rates are approximately 30 percentage points higher for ‘Control group’ firms. In contrast, for ‘Business related services’ only small differences can be found. For only one industry group (‘Recreation/Sports/Culture/Others’) ‘On-incubator’ firms have higher survival rates than ‘Control group’ firms (Fig. 4).

Fig. 4

Comparison of survival rates according to firms’ industry affiliations (aggregated analysis). Note: Industry groups ‘Construction’ and ‘Education’ are omitted due to very small sample sizes

5.2 Market exit dynamics

In analyzing market exit dynamics, it is of particular interest how hazard rates change in the first years after the market entry. Do incubators provide a supporting business environment to reduce the comparable high failure risk start-ups face? The subsequent analyses are based on closure rates including those firms that were subject to M&A. Previous research showed that analyses of (incubator) firm survival and exit dynamics are remarkably robust if M&A-cases were treated either as survivors or closures, or if they would be completely excluded from the analysis (Schwartz 2009). Six firms were omitted in the analysis, since an exact date of closure could not be identified.

Table 5 gives the Kaplan–Meier estimates of the survivor function (Survival), the number of firms at risk (Risk), the number of firm closures (Deaths) and the hazard rates (Hazard). For comparative purpose, we report these measures for the aggregated sample as well. Additionally, Fig. 5 provides the graphical representation of both survival curves and hazard rates for ‘On-incubator’ firms and ‘Control group’ firms according to the five incubator locations. As mentioned in Sect. 4.3, survivor function give the probability of surviving the respective interval, whereas hazard rates specify the instantaneous risk of firm closure conditional of having survived until the beginning of this interval. Analysis time as well as graphical representation is restricted to 10 years, because of decreasing number of observations in the time intervals with simultaneously increasing observation time.

Table 5 Life tables for ‘On-incubator’ firms versus ‘Control group’ firms

Fig. 5

Kaplan-Meier survival curves (left columns) and hazard estimates (right columns) for ‘On-incubator’ firms versus ‘Control group’ firms according to incubator location

First, with respect to the relationship between firm age and the development of hazard rates, there is no evidence for the existence of a specific high-risk period confronting either ‘On-incubator’ firms or ‘Control group’ firms with high risks of firm closure. We do not find evidence for high hazard rates immediately after market entry (‘liability of newness’).

Second, findings of Table 5 and Fig. 5 reinforce the results of the previous section with respect to the two groups of incubator locations. For the TGZH and the TZD, only slight differences between ‘On-incubator’ firms and ‘Control group’ firms for the probability of surviving a period of 10 years are revealed. For the TGZH, for instance, the survivor functions show nearly equivalent probabilities of surviving the ten-year period after start-up for ‘On-incubator’ firms (76.6 %) and ‘Control group’ firms (77.1 %). Accordingly, a log-rank test of equality of survivor functions does not indicate a statistical significant difference on the 10 %-level between the two samples. Moreover, for these two incubator locations, the comparison of hazard rates between ‘On-incubator’ firms and ‘Control group’ firms does not indicate the presumed risk-reduction in the first years after the market entry.

Third, with respect to the remaining incubator locations (TIPJ, TIGN, RIGZ), log-rank tests indicate statistical significant lower survival chances for ‘On-incubator’ firms compared to ‘Control group’ firms. The most striking difference is found for TIPJ. Whereas for incubated firms, the probability of surviving 10 years after start-up is 53.2 %, non-incubated firms have a survival probability of 87.8 %. Furthermore, the hazard rates for TIPJ ‘Control group’ firms are quite low (mean hazard rate of 1.3 % over the entire ten-year observation period) and for every interval considered below the respective hazard rate of ‘On-incubator’ firms (mean hazard rate of 6.7 %). The respective graphs in Fig. 5 give a good impression of these patterns.

6 Conclusions

This paper provides insights on the contribution of incubation towards long-term firm survival. Therefore, a sample of 371 start-up firms supported by five German publicly-funded incubation initiatives is tracked over time and is contrasted with the development of an appropriate control group of 371 comparable start-ups not receiving support by an incubator. We analyze and compare survival rates as well as the evolution of the risk of market exits.

We do not observe that firms located in incubators have higher survival chances than comparable firms located outside those incubator organizations. For neither of the five incubator locations investigated, we find statistically significant higher survival probabilities for ‘On-incubator’ firms compared to ‘Control group’ firms. Nevertheless, for three incubator locations the analysis even reveals statistically significant lower chances of survival for those start-ups receiving support by an incubator. Our somewhat disillusioning results seem to confirm existing concerns raised previously (Elle et al. 1997; Pleschak and Werner 1999; Seeger 1997; Westhead and Storey 1994). Our study does not show that being located in an incubator significantly increases the chances of long-term business survival. The empirical analysis of this paper, therefore, raises some doubts regarding the impacts of incubation on long-term firm survival.

This study has some limitations that must be addressed. Most importantly, matching of incubator firms with non-incubated firms to control for selection bias is performed along four key dimensions, i.e. location, industry, age, and legal form. Other explanatory variables that capture important characteristics of both samples and that might contribute to survival could not be included in the matching process, such as the influence of team members (Ensley et al. 2002), human capital endowment (Acs et al. 2007), innovation and patent activity (Cefis and Marsili 2006; Wagner and Cockburn 2010), founding conditions (Geroski et al. 2009) or network embeddedness of start-ups (Shane and Stuart 2002). Most in-depth characteristics can not be observed ex-post for discontinued firms since they had to be measured at firm foundation or close to start-up. A desirable approach could be the construction of a control group of firms that did apply for incubation but ultimately did not become incubated, as proposed by Sherman and Chappell (1998). Such research design, unfortunately, is not practicable in Germany, because incubator management does not maintain such information.

In the present study, the impact of incubator support on incubator firm survival is roughly measured through the fact that firms have been located on incubators. There is no detailed firm-specific data available about what actually happens during the stay in the incubator. For instance, we could not account for incubator-specific intensity of managerial support. In this respect we want to emphasize that this study does not show what actually would have happened to the incubator firms (‘On-incubator’-sample) if they had not been incubated. For instance, key entrepreneurial learning processes probably would have never occurred or valuable networks never would have been established. It is not possible to assess how much these firms were supported by the incubation process.

Also, there exists no standardized incubator organization, that is no two incubators are alike (Allen and McCluskey 1990; Sherman and Chappell 1998). Evaluation of outcomes suffers from this lack of homogeneity, as already detailed in Sect. 2.3. Incubators are mostly tailored to more or less specific regional circumstances and to characteristics of their target group, and are expected to fulfill different roles in regional innovation systems. Whereas in so called high-tech regions the commercialization of academic research might be the primary incubator objective, for incubators located in economically depressed regions, the focus might be more on general economic development processes. Research could therefore focus on similar studies that compare differences in the contribution to firm survival according to the type of incubator, such as diversified versus more specialized incubators (Aerts et al. 2007; Schwartz and Hornych 2010) or profit versus non-profit incubators (Grimaldi and Grandi 2005). This could help to answer the question which incubation model is most suitable to increase survival chances.

Appendices

Appendix 1

See Table 6.

Table 6 Correlation matrix; N = 43,467

Appendix 2

See Table 7.

Table 7 Results of probit estimations for the probability of receiving public support by an incubator organization (standard errors in parentheses)

Appendix 3

See Table 8.

Table 8 Comparison of the distribution of main matching variables for ‘On-incubator’ firms, potential controls before matching, and ‘Control group’ after matching; comparison per incubator location