Tobias Ebert, Thomas Brenner, Udo Brixy. Citation and published article at Springer: New Firm Survival: The Interdependence between Regional Externalities ...
New Firm Survival: The Interdependence between Regional Externalities and Innovativeness
Tobias Ebert, Thomas Brenner, Udo Brixy
Citation and published article at Springer: New Firm Survival: The Interdependence between Regional Externalities and Innovativeness Note: This is not the final, proofread article. Please cite the published article.
ABSTRACT Many studies have shown that regional externalities play a crucial role for the survival prospects of newly founded companies. Recent research moreover provides evidence that not all businesses are affected by these externalities in the same way. Relying on new and representative panel survey data, which contains information about 6,776 German newly founded firms from nearly all economic sectors between 2007 and 2011, this paper suggests that the effects of regional externalities on survival are contingent upon the start-up’s innovative behavior. First, we show that introducing market novelties is not necessarily beneficial for newly founded firms, but might even endanger their survival. Second, we find that being located in spatial proximity to similar firms is important for start-ups in non-high-tech environments and unfolds a positive influence on survival only for less innovative companies. On the contrary, high-tech start-ups are rather affected by a diverse economic structure. Thereby, we suggest that the recombination of knowledge from diverse sources can lead to overly complex innovation projects with detrimental effects on the start-up’s survival prospects. Keywords: Firm survival, Innovation, Externalities JEL-Classification: D22, L26, O33,
1 Introduction The level of entrepreneurial activity has long been a major topic for policymakers. Researchers have often focused on the determinants of new firm emergence (e.g. Armington and Acs 2002; Fritsch and Falck 2007). While these studies contribute to explaining regional variations in entrepreneurship, they say little about the processes that affect the success of start-ups in the phase after their foundation. In fact, the failure rate among young firms is very high: around 50% of all new manufacturing plants close down within the first five years, and only 20% survive longer than ten years (Mata and Portugal 1994; Knaup 2005). Hence, “the survival or success of new firms is more essential to a regional economy than merely the presence of a large number of new firms” (Schutjens and Wever 2000: 136). Consequently, many researchers have begun to investigate what determines the survival prospects of newly founded firms. Although many studies with various focuses have been undertaken in this area (e.g. Audretsch 1991; Brixy and Grotz 2007; Brüderl et al. 1992; Cefis and Marsili 2006, 2011; Geroski et al. 2010; Renski 2011; Van Praag 2003), the mechanisms underlying new firm survival are still far from being understood. We contribute to clarifying this issue in two ways. First, recent research (Pe’er and Keil 2013) suggests that regional externalities do not affect all start-ups in the same way. Instead, the impact of externalities on start-ups is moderated by certain company characteristics. So far, only few company attributes and only certain types of regional externalities have been considered (Puig et al. 2014; Renski 2015; Howell et al. 2016; Pe’er et al. 2016). Our study is the first to analyze the extent to which the innovation behavior of start-ups moderates how much they can benefit from regional externalities of localization and economic diversity. Second, studies on the relationship between regional externalities and new firm survival focusing on Germany are rare so far, so we check previous empirical findings for a large number of German start-ups. The study relies on new and representative panel survey data containing information about 6,776 German firms from almost all economic sectors that were started between 2007 and 2011. These data yield valuable insights into the first critical years of the analyzed start-ups. By applying a semiparametric Cox regression, the study is able to demonstrate that start-ups introducing national or global market novelties are not benefiting, but might even suffer from being located in spatial proximity to similar firms. In contrast, highly innovative start-ups appear to be affected by a diversified economic environment. In this respect, the study adds knowledge to both agglomeration and entrepreneurship research by deepening the understanding of the relationship between agglomeration, innovation and new firm survival. 1
To achieve these goals, the study is structured as follows: first, the theoretical background on the types and sources of agglomeration externalities is presented. Subsequently, a review of the relevant literature is provided. In section three, hypotheses are derived. The data used and the statistical approach applied are described in section four. The results of the statistical analysis are presented and discussed in section five. Section six discusses the robustness and limitations of our findings, while the final chapter concludes.
2 Theory and empirical evidence 2.1 Theoretical underpinnings of agglomeration externalities “Regardless of the country, spatial concentration of aggregate activity is a fact of the economic landscape” (McCann & Folta 2008: 533). One reasonable explanation for this tendency to concentrate lies in the assumption that firms benefit in some way from being located in proximity to each other. These forces are dynamically self-intensifying and depend on the number and sector of establishments in a region. The more firms are already concentrated in a region, the stronger are the external effects on productivity, thereby attracting more employment, which in turn leads to further agglomeration. These effects are external to the firms, meaning that every establishment in a region can benefit from them (cf. Scitovsky 1954; Fujita and Thisse 1996). Agglomeration externalities (which are understood here as any kind of external economies arising from the regional economic structure) can be broadly split into externalities arising from an accumulation of similar firms and externalities resulting from a diverse economic structure (McCann and Folta 2008). An accumulation of similar organizations (also termed localization externalities henceforth) are often called “MAR” externalities, after Marshall (1920), Arrow (1962) and Romer (1986). These externalities need to be strong enough to offset potential adverse effects of congestion - such as high rent levels and large amounts of traffic - and increased competition due to strong concentrations of similar firms (Folta et al. 2006; Prevezer 1997; Schmalensee 1978). In general, the literature explains spatial concentrations by pointing out that companies in close proximity to one another not only experience increased competition, but also benefit from superior access to specialized labor, specialized inputs, technology spillovers as well as greater demand (McCann and Folta 2008). Jacobs (1969) assumes that regional externalities cannot only arise from an accumulation of similar firms, 2
but that knowledge generated in one industry may also be adopted in another industry. These externalities arising from a diverse regional economic structure are often termed ‘Jacobs externalities’. It is important to bear in mind that Jacobs and localization externalities are not mutually exclusive. On the one hand, a region with a specialization in a certain industry might well also possess a diverse economic structure in other sectors. On the other hand, it seems fruitful to assume that different kinds of spillovers emerge from different agglomeration economies (Tödtling et al. 2009). Localization externalities are expected to lead to incremental innovations and process innovations that make production more efficient. This type of externalities spurs price competition and is important for sustaining regionally well-established industries (Henderson et al. 1995). In contrast, externalities of regional economic diversity are expected to facilitate genuine new combinations of knowledge, which supposedly foster the invention of new products or services, thereby creating new markets and providing opportunities for employment growth. However, these spillovers between industries only occur efficiently if complementarities between industries exist and if the cognitive distance between sectors is not too large (cf. Nooteboom et al. 2007). Frenken et al. (2007) try to incorporate these preconditions. They introduce a finer subdivision of the term ‘diversity’ by differentiating between related and unrelated diversity, whereby spillovers “occur primarily among related sectors and only to a limited extent among unrelated sectors” (Frenken et al. 2007: 688). To determine whether industries or technologies are related, one widely used possibility is to apply an entropy measure which implies that industries belonging to a common industrial classification are related (e.g. Boschma and Iammarino 2009; Frenken et al. 2007). Other approaches for assessing the relatedness of industries or technologies include, for example, patterns of patent citations (e.g. Rigby 2015) or product portfolios (e.g. Neffke and Henning 2008).
2.2 Evidence on new firm survival Agglomeration externalities and the survival chances of new firms The theoretical framework has outlined the differentiation between agglomeration externalities that arise either due to an accumulation of similar firms or to a diversified economic structure. Neffke et al. (2012) show that young companies are affected by these agglomeration effects in a different way to established firms. In the following we therefore discuss the studies that deal with new or young companies. Wennberg and Lindqvist (2010) present an overview of early studies investigating the effect of localization externalities on new firm survival. They arrive at conflicting results because these studies 3
use “different levels of geographical aggregation and different measures of agglomeration [as well as] different levels of industry aggregation” (Wennberg and Lindqvist 2010: 225). Weterings and Marsili (2015) provide an explanation for the so far heterogeneous findings. By applying a competing risk model they are able to show that a spatial concentration of industries lowers the probability of exit due to failure and increases the likelihood of exit due to merger & acquisition (M&A). Studies that take different modes of exit into account concordantly find a positive relationship between localization externalities and new firm survival. Thereby, these effects of localization externalities appear to be primarily relevant in traditional and low-tech sectors and seem to be stronger for relative agglomeration measures which are depicted on a broader geographical scope (Renski 2011; Wennberg and Lindqvist 2010; Weterings and Marsili 2015). Recent research also shows that the influence of regional externalities might be moderated by individual company characteristics. Interaction effects have been found for endowment with assets and human capital (Pe’er and Keil 2013), internationalization activities (Puig et al. 2014), the founder’s/owner’s industry experience (Renski 2015), subsidies received (Howell et al. 2016) or company growth patterns (Pe’er et al. 2016). Only very few studies have examined the link between regional economic diversity and new firm survival, and the existing findings are mixed at best. Renski (2011) provides the first study that explicitly links new firm survival to regional economic diversity. By measuring how far a regional industry structure differs from the national composition, it is shown that regional industrial diversity generally increases survival chances for new firms in the USA. This relationship is particularly pronounced for knowledgeintensive start-ups and exhibits a stronger impact when it is depicted on a broader geographical scope. In contrast, Howell et al. (2016) show that a region’s overall economic diversity is not related to new firm survival in China. However, they find that related variety has a beneficial impact on survival, while unrelated variety has a detrimental impact. Furthermore, the influence of these diversity measures is moderated by subsidies received, in the sense that heavily subsidized firms are generally affected to a lesser degree by measures of regional diversity. Finally, Basile et al. (2016) find for Italy that related variety enhances survival prospects only in manufacturing industries, while unrelated variety only has a positive influence in service industries.
Innovative activities and new firm survival There is quite a broad supply of literature on the effects of innovative behavior on the success of small 4
and medium-sized companies (see Rosenbusch et al. 2011 for a meta-analytic review). However, these studies mainly use growth-related performance measures, thereby permitting only indirect conclusions in terms of company survival. Conversely, the existing studies that directly link innovative activities to company survival often include companies of all ages and do not take into account differences between entrepreneurial and established firms (Buddelmeyer et al. 2010; Fontana and Nesta 2009). Only a few studies directly link innovative activity to company survival and examine start-up businesses only or separately. Thereby, the innovative behavior of firms is measured in different ways. One approach focuses on the use of intellectual property rights (Helmers and Rogers 2010; Jensen et al. 2008). Another common empirical strategy differentiates between more or less innovative start-ups on the basis of questions in surveys about the innovativeness of products or the technology used. In an early surveybased study of Dutch manufacturing companies, Cefis and Marsili (2006) arrive at the conclusion that in low-tech industries, young firms’ survival benefits from innovation, while for high-tech industries no such innovation premium is observable. Thus, being an innovator can even increase the risk of failure in the longer term. This potentially endangering effect of innovations is supported by findings showing a negative relationship between the status of being an innovator and survival prospects for newly founded firms in France and Finland (Boyer and Blazy 2014; Hyytinen et al. 2015). By applying a competing risk model that disentangles exit due to failure and exit due to merger or acquisition, Cefis and Marsili (2011) are able to show for Dutch manufacturing companies that non-innovative companies in low-tech industries, as well as innovating companies in high-tech industries, have the highest risk of failure. This leads them to conclude that in low-tech industries, being innovative is sufficient for survival, while in the fast-changing environment of high-tech industries, being innovative may only represent the entry point to competition. As innovation is the common denominator of these high-tech companies, there is a need to outperform competitors with highly innovative and risky products.
3 Hypotheses As outlined in the introduction, we focus on two aspects: the impact of agglomeration externalities on new firm survival and the interaction between agglomeration effects and innovation. More recent studies (Renski 2011; Wennberg and Lindqvist 2010; Weterings and Marsili 2015) tend to show a beneficial impact of localization externalities on new firm survival, which seems more prevalent in low-tech 5
environments. No studies have yet been conducted on the effects of localization externalities for Germany. Nevertheless, we expect to confirm the results obtained for other countries: H1a: Regional localization externalities exert a positive influence on new firm survival in non-hightech environments. Insights into the influence of economic diversity on new firm survival are very sparse and mixed. So far, the few existing findings suggest that regional diversity increases the probability of company survival. Consequently, one important aim of this study is to validate this potential relationship between regional diversity and new firm survival by testing the following hypothesis. H1b: Regional diversity exerts a positive influence on new firm survival. Recent studies show that the impact of agglomeration externalities depends not only on the technological environment, but also on individual company characteristics (Puig et al. 2014; Renski 2015; Howell et al. 2016; Pe’er et al. 2016). So far, only few different company attributes have been tested as potential moderators of localization externalities. To the authors’ knowledge, no study has tested the extent to which localization effects are moderated by the innovative behavior of the start-up. The moderating role of company attributes for the effects of regional economic diversity on new firm survival has almost not been studied at all (Howell et al. 2016). We measure innovative behavior in terms of market novelties, which are associated more with radical rather than incremental innovations. Intra-industry spillovers – a first aspect of localization externalities – are assumed to lead to incremental product and process innovations (Tödtling et al. 2009; Frenken et al. 2007), so they should conceptually not be related to our innovation measure. Regarding the remaining aspects of localization externalities, such as local labor and suppliers, Pe’er and Keil (2013) show that the relationship between localization externalities and new firm survival is contingent upon a start-up’s endowment with tangible and intangible assets. To be more precise, they find that localization benefits are stronger for companies with little asset endowment in comparison to their competitors. Being in possession of a market novelty can be understood as a form of intangible asset endowment, and the following hypothesis can be derived: H2a: The survival of start-ups introducing market novelties is less positively affected by localization externalities than the survival of start-ups without such novelties. 6
Externalities of a diverse regional economic structure are assumed to arise from inter-industry spillovers, which lead to a recombination of knowledge from different sectors and, thus, to rather radical product innovations (Tödtling et al. 2009). In the light of findings suggesting that engaging in complex and uncertain innovations can endanger the survival prospects of start-ups (Cefis and Marsili 2011), it seems legitimate to assume that recombining diversified knowledge in innovation projects reduces survival chances. Accordingly, if inter-industry spillover effects are actually present, the interaction between regional economic diversity and the introduction of market novelties should exhibit a negative impact on the survival chances of newly founded businesses: H2b: The survival of start-ups introducing market novelties is less positively affected by regional economic diversity than the survival of start-ups without such novelties.
4 Data and methodology 4.1 Data We use data from the Mannheim Start-up Panel (previously known as the KfW/ZEW Start-up Panel) of the Centre for European Economic Research (ZEW). The data cover annual surveys conducted in 6,776 firms that were founded between 2007 and 2011. 1,074 closing events were observed between 2008 and 2012. The panel covers almost all industries and is representative of the whole of Germany, excluding only agriculture, mining and quarrying, electricity, gas and water supply, health care and the public sector. The sample is stratified according to two criteria: start-ups that received financial support from the KfW banking group and start-ups in the high-tech sector 1 are oversampled. The parent population (the Mannheim Enterprise Panel of the ZEW) is maintained by the ZEW in cooperation with Creditreform, the largest business information service in Germany. The entities are legally independent firms (thus excluding de-merger foundations or subsidiaries) which are run by at least one full-time entrepreneur. The survey data are collected via computer-aided telephone interviews with those involved in the management of the newly founded businesses (Fryges et al. 2009). A survivor bias could exist in the first two cohorts, which consist not only of firms started up in the previous year, but also of firms that had existed up to three years before the first interview. To avoid this bias, the panel was reduced to include
1
The definition of high-tech industries follows the approach of Legler and Frietsch (2006).
7
only those companies that were founded in the year immediately preceding their first interview. Additionally, to ensure a sufficient number of observations for all periods of analysis, the maximum observation period for each company is set to four years, meaning that the 2007 cohort will only be monitored until the wave of 2011.
Survival as the dependent variable The aim of the analysis is to explain the time that elapsed between the entry and the exit of the company. In this respect, the panel possesses the advantage that the Creditreform database makes it possible to identify the precise month when a company actually started to actively participate in business life (e.g. by renting business rooms or taking out a loan). This definition of entry based on active participation in business life should realistically represent the point in time from which a start-up is actually subject to a mortality risk. Accordingly, this is an advantage of our data, as this definition of entry should be more realistic than previously applied definitions of entry, such as enrolment in business registers (e.g. Weterings and Marsili 2015) or hiring the first employee (e.g. Renski 2011). In terms of company exit, a common problem with empirical research is that little information is available for subjects that have left the panel. It is therefore difficult to distinguish whether a firm is no longer in existence, or whether the subject has simply changed their contact details or such like. Another advantage of the present data is that they rely on two independent sources regarding firm closures. Besides information regarding a firm’s closure that is obtained from interviews, some firms are recognizable as closed according to an identifier within the Creditreform database. Additionally, for companies that no longer responded but were not labeled as closed, Creditreform directly researched their status. However, information regarding closure that is obtained from identifiers and research might only be available with a considerable lag. Hence, there may be a period of unmonitored time between the last interview and the date of closing (Fryges et al. 2009). In this analysis, an unobserved period of one year is accepted; otherwise the company is labeled as right-censored at the time of the last interview. Finally, previous research revealed that different types of company exit from the market might represent different economic outcomes. However, in the present data it is not possible to distinguish between different exit modes. This naturally reduces the significance of the results and needs to be carefully considered when interpreting the outcomes.
8
Variables representing agglomeration externalities Agglomeration externalities are depicted at the level of 258 labor market regions (BBSR 2017). We favored this level over more fine-grained levels for two reasons: first, various authors have found that the effects of agglomeration externalities on new firm survival are stronger when a broader regional scope is applied (Renski 2011; Wennberg and Lindqvist 2010). Second, labor market regions form functional rather than administrative entities. Their delineation follows daily commuting patterns, reflecting the regional scale for regular interactions. Accordingly, spatial spillover effects should be unlikely to reach beyond these boundaries (Eckey et al. 2006). As for the classification of industries, all of the indicators presented subsequently are based on a three-digit NACE classification (Federal Statistical Office 2008). All the indicators are based on data from the Establishment History Panel of the Institute for Employment Research (IAB), which covers all establishments with at least one employee in Germany (Gruhl et al. 2012). While the start-ups are observed in the period from 2007 to 2011, the agglomeration externalities represent averages for the period 2003-2007. Measuring agglomeration externalities in a period before the observation of the dependent variable is relevant for two reasons. First, agglomeration externalities typically exert their influence more in the medium and long run (Cainelli et al. 2014; Glaeser et al. 1992). Second, measuring agglomeration externalities prior to the founding of the observed start-ups excludes endogeneity that may arise because the foundation and survival of new firms naturally also alters the regional economic structure. Consequently, if survival and agglomeration externalities were measured in the same period, the direction of the relationship between the variables would not be unambiguous. To operationalize localization externalities, previous survival studies generally suggested a superiority of relative quotients over absolute measures (Renski 2011; Weterings and Marsili 2015). To capture localization externalities, an important distinction can be made between regional concentration and regional specialization (Sternberg and Litzenberger 2004). In this respect, an industry is regarded as being concentrated in a region when, in relation to the entire nation, this industry features a high spatial density in that region. To measure regional concentration, we divide the regional number of employees in a given industry by the area of that region (in km2) and then normalize this value using the spatial density of that industry in the entire nation. In contrast, a region is regarded as being specialized in an industry when, in relation to the entire nation, a large share of the region’s economic activities takes place in a given industry. To measure regional specialization, we apply the standard localization quotient. That is, we divide the regional number of employees in a given industry by the total number of employees in 9
that region and then normalize this value using the employment share of that industry in the entire nation. 𝑒𝑖𝑟 𝑎 𝐶𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛𝑖𝑟 = 𝑒𝑟 𝑖 𝑎
(1)
𝑒𝑖𝑟 𝑒 𝑆𝑝𝑒𝑐𝑖𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛𝑖𝑟 = 𝑒𝑟 𝑖 𝑒
(2)
where (i) denotes the respective industry, (r) the respective region, (e) the number of employed people and (a) the area size. To capture regional economic diversity, we first employ the reversed score of the Krugman index. Following Dauth (2013), this index shows how far the regional industry mix deviates from the mix of the whole nation: 𝑁
𝐷𝑖𝑣. 𝐾𝑟𝑢𝑔𝑚𝑎𝑛𝑖𝑟 = − ∑ | 𝑖`=1,𝑖`≠𝑖
𝑒𝑖`𝑟 𝑒𝑖` − | 𝑒𝑟 𝑒
(3)
where (eir) is the employment in industry (i) and region (r). The more specialized a region is, the more negative the index becomes. As an additional measure of diversity, we apply the reversed score of the Herfindahl index. The Herfindahl index depicts the composition of the regional industry mix by summing the squared employment share of all three-digit sectors in a region: 𝑁
𝑒𝑖𝑟 2 𝐷𝑖𝑣. 𝐻𝑒𝑟𝑓𝑖𝑛𝑑𝑎ℎ𝑙𝑟 = − ∑ ( ) 𝑒𝑟
(4)
𝑖=1
where (eir) is the employment in industry (i) and region (r). The more the regional industry mix is shaped by a few dominant industries, the more negative the index becomes. Finally, to explore regional economic diversity in greater depth, we split diversity into related and unrelated variety using the standard entropy measure suggested by Frenken et al. (2007). If all the threedigit shares fall under one two-digit sector (SG) (where (g) =1, …, (G)), the two-digit shares (Pg) can be derived by summing the shares of all the three-digit industries (pi) belonging to a two-digit sector: 10
(5)
Unrelated Variety (UV) is the entropy at the two-digit level:
(6)
Related Variety (RV) is the weighted sum of the entropy within each two-digit sector:
(7)
(8)
where:
In total, we employ six different measures of regional externalities, whereby two aim at capturing localization externalities and four aim at capturing regional economic diversity. As the correlation matrix (table 1) shows, the two localization indices (Concentration and Specialization) are only weakly correlated with each other (.24) indicating that regional concentration and regional specialization are two different constructs. The Krugman and Herfindal indices of diversity are moderately correlated (.52), and hence both possess a unique share of variance. Both, the Krugman and the Herfindahl index, are moderately correlated with Related Variety (.54) and strongly, but not perfectly, with Unrelated Variety (>.75). The measures of Related and Unrelated Variety are moderately correlated with each other (.42). Table 1. Correlation matrix of agglomeration measures. Concentration Concentration
1.000
Specialization
KrugmanDiversity
HerfindahlDiversity
Related Variety
Specialization
0.242
1.000
Krugman-Diversity
0.205
0.071
1.000
Herfindahl-Diversity
0.091
0.044
0.522
1.000
Related Variety
0.099
0.034
0.539
0.542
1.000
Unrelated Variety
0.229
0.117
0.798
0.749
0.421
Unrelated Variety
1.000
11
Innovation variables Following the OSLO manual (OECD 2005), all the participants of the Mannheim Start-up Panel are asked every year whether they introduced a market novelty and if so, whether this market novelty is new to the region, new to Germany or new to the world. We calculate a binary variable for each novelty degree, whose value changes from zero to one in the year in which the novelty was introduced and then remains at this value. The three categories are mutually exclusive meaning that, for example, a global novelty – which is naturally also new to the region and the nation – does not simultaneously count as a regional or national novelty. This operationalization can be understood as indicating whether the company possesses a market novelty of a certain novelty degree in its portfolio and allows us to examine the unique effect of each novelty degree on survival. This study is the first to be able to a) represent innovative behavior by means of a time-varying covariate and b) additionally differentiate innovations according to their degree of novelty.
Control Variables Our data permit manifold control variables that proved to be of relevance in previous empirical studies. They include the founder’s gender, nationality, experience and qualification-level (van Praag 2003) and the current firm size (Aldrich and Auster 1986). Furthermore, we control for public funding received2 (Désiage et al. 2010), industry sectors (Audretsch 1991) cohort effects (Singh and Lumsden 1990; Strotman 2007), population density (Falck 2007; Stearns et al. 1995) and East-West differences (Wyrwich 2013). Finally, we also control for the start-up’s legal form by distinguishing whether the company entered the market as a capital company (Kapitalgesellschaft) or not (Harhoff et al. 1998). Operationalizations of these variables can be found in table 2. The metric variables for industry experience and current employment are transformed to their log values, as this approach is preferred by some previous empirical studies (Brüderl et al. 1992) and lower AICs in our analyses. The data were additionally divided into a high-tech and a non-high-tech group (see table 2). Regressions are also run for these categories separately.
2
However, the results for this variable may not be reported as they are subject to data protection regulations.
12
4.2 Descriptive analysis Table 2 contains descriptive statistics for all the independent variables adjusted for sampling weights. It is indicated whether a potential change of the variables over time is explicitly considered. All the spatial variables refer to the region in which the business was founded, although a very small proportion of startups change their location during the observation period. This is controlled for in an unreported insignificant control variable. The overrepresentation of firms belonging to high-tech industries allows us to conduct separate analyses for these economically crucial sectors. Regarding the innovation variables, table 2 shows that innovations leading to market novelties are not very common and are unequally distributed across high-tech (HT) and non-high-tech (NHT) sectors.
13
Table 2: Descriptive statistics of all independent variables (weighted values)
Descriptive statistics of independent variables Variable
Mean
Description Sample:
SD
Min
Max
Time
FULL
HT
NHT
FULL
HT
NHT
FULL
HT
NHT
FULL
HT
NHT
Variance
Invariant
Continuous variables Experience_Log
Log of highest experience in industry at founding
2.167
2.383
2.148
1.045
0.86
1.057
0
0
0
3.912
3.912
3.912
Employ_Log
Log of current employment in full time equivalents
0.781
0.773
0.782
0.803
0.78
0.805
0
0
0
6.1
4.85
6.1
Variant
Pop_Density
Population density
603.9
678.4
597.5
803.8
872.7
797.2
38.1
38.1
38.1
3927.9
3927.9
3927.9
Invariant
Concentration
Spatial density of industry in relation to national density
3.356
3.924
169.78
6.302
5.569
6.356
0
0
0
169.78
70.5
169.78
Invariant
Specialization
Employment share of industry in relation to national share
8.795
5.294
9.099
64.74
31.19
66.86
0
0
2553
522.15
2553
Invariant
Krugman-Diversity
Deviation from national industry mix
-0.517
-0.503
-0.516
0.113
0.112
0.113
-1.107
-1.107
-1.107
-0.318
-0.323
-0.318
Invariant
Herfindahl-Diversity
Sum of squared employment shares in region
0.197
0.194
0.197
0.011
0.129
0.011
0.013
0.013
0.013
0.268
0.268
0.268
Invariant
Related Variety
Weighted sum of entropy within two-digit sectors
1.206
1.206
1.206
0.091
0.086
0.091
0.782
0.782
0.782
1.365
1.365
1.365
Invariant
Unrelated Variety.
Entropy at two-digit industry level
5.232
5.246
5.231
0.198
0.205
0.198
3.078
3.078
3.078
5.508
5.508
5.508
Invariant
KFW-funding
Promoted by the KfW banking group
0.060
0.027
0.063
Industry_Sector1
Cutting-edge manufacturing (high-tech)
0.004
0.048
Invariant
Industry _Sector2
High-tech manufacturing (high-tech)
0.004
0.049
Invariant
Industry _Sector3
Technology-intense services (high-tech)
0.055
0.689
Invariant
Industry _Sector4
Software & consultancy (high-tech)
0.017
0.214
Industry _Sector5
Non-high-tech manufacturing (non high-tech)
0.044
0.048
Invariant
Industry _Sector6
Skill-intense services (non high-tech)
0.073
0.080
Invariant
Industry _Sector7
Business-oriented services (non high-tech)
0.118
0.129
Invariant
Industry _Sector8
Consumer-oriented services (non high-tech)
0.354
0.385
Invariant
Industry _Sector9
Construction (non high-tech)
0.125
0.136
Invariant
Industry _Sector10
Wholesale and retail trade (non high-tech)
0.205
0.222
Invariant
Cohort_2007
Entry to panel 2007
0.228
0.225
0.229
Invariant
Cohort_2008
Entry to panel 2008
0.281
0.260
0.283
Invariant
Cohort_2009
Entry to panel 2009
0.226
0.246
0.225
Invariant
Cohort_2010
Entry to panel 2010
0.164
0.160
0.165
Invariant
Cohort_2011
Entry to panel 2011
0.100
0.108
0.099
Invariant
German_Involved
German involved in founding
0.907
0.961
0.902
Invariant
Women_Only
Only women involved in founding
0.241
0.119
0.252
Invariant
Quali_Academic
Founder(s) graduated from university or college
0.360
0.612
0.338
Invariant
Legal_Form
Entered market as capital company
0.350
0.557
0.332
Invariant
East_West
Located in East Germany
0.211
0.183
0.214
Invariant
Regional_Novelty
Regional market novelty in product portfolio
0.089
0.084
0.090
Variant
National_Novelty
National market novelty in product portfolio
0.097
0.147
0.092
Variant
Global_Novelty
Global market novelty in product portfolio
0.041
0.108
0.036
Variant
13,899
5,293
8,606
Dummy variables
N annual spells
Invariant
Invariant
14
4.3 Methodology A semiparametric Cox regression (Cox 1972) is used to test the hypotheses. The advantages of this widely used approach lie in its flexibility and the robustness due to an absence of distributional assumptions. Graphical examinations as well as a Grambsch-Therneau Test (Grambsch and Therneau 1994) indicate no violations of the preconditioned proportional hazards assumption in any model specification (see P.H.-Test in table 3). Within the model, the Breslow approximation for tied failures is applied. We additionally control for potential unobserved within-group correlation among start-ups in the same labor market region by fitting a model with shared group-level frailty. Accordingly, the hazard function for subject (j) in region (i) then reads as:
hijt = h0(t) 𝛼i exp(β0 + xj βx)
(9)
where (ho) is the so-called baseline hazard and exp() is taken to ensure that hijt () cannot become negative, (t) is time, (xj) is a row vector of multiple predictors, (βx) is a column vector of regression coefficients and (αi) represents the unobservable positive quantities. These random effects are assumed to follow a gamma distribution with a mean of 1 and a variance of θ, which is estimated from the data. For ease of interpretation, all the metric variables are z-standardized and the coefficients are reported in exponentiated form, so that they represent hazard ratios. Tests for multicollinearity among independent variables using the variance inflation factor revealed no problems (VIF < 2 for all variables across all model conditions).
5 Results To check the proposed hypotheses, regressions are run for the full sample of all 6,776 start-ups, as well as for 2,588 high-tech and 4,188 non-high-tech start-ups separately (see results in table 3). First, we run a baseline model without any interactions between measures of innovation and agglomeration (models 1-3). Subsequently, we run models including one measure of localization and one measure of diversity in the model. In other words, we never include two localizations or two or more diversity measures at the same time, because especially the diversity measures are related to each other (see table 1). All the localization and diversity measures are examined, but we report only the results for those analyses in which interactions between innovation and agglomeration measures exhibit a significant outcome. The 15
majority of significant findings are obtained for the interactions of our innovation measures with the measures Concentration and Krugman Diversity (models 4 -6). Mainly insignificant results are obtained for the other measures, so the regression results are reported in appendix 1 and the few significant results are reported in the text. For all the models, the hazard rate is low for the initial phase, then rises sharply and peaks after about one and a half years, with a tendency to decrease thereafter. This pattern supports the idea of a liability of adolescence and a honeymoon phase after the founding (Fichman and Levinthal 1991; Brüderl and Schüssler 1990). Table 3: Results of the semiparametric Cox regression.
Model Condition
(1) Full Sample
Baseline Model (2) High-Tech
HR p-value HR Cohort controls YES YES Sectorial controls YES YES KfW funding control YES YES German_Involved 0.706 0.790 (0.040)* Quali_Academic 0.947 (0.452) 0.972 Women_only 1.277 1.191 (0.021)* Experience_Log 0.869 (0.000)*** 0.830 Employ_Log 0.884 (0.001)** 0.812 Legal_Form 0.526 (0.000)*** 0.438 Regional_Novelty 1.038 (0.707) 0.990 National_Novelty 0.941 (0.564) 1.200 Global_Novelty 1.229 (0.099) 1.337 East_West 0.862 (0.116) 0.760 Pop_Density 1.070 (0.261) 1.001 Concentration 0.920 (0.203) 0.990 Krugman-Diversity 0.986 (0.697) 0.980 Reg_Nov#Conc. Nat_Nov#Conc. Glob_Nov#Conc. Reg_Nov#Krugm.-Div. Nat_Nov#Krugm.-Div. Glob_Nov#Krugm.-Div. N subjects 6,776 N annual spells 13,899 N exits 1,074 Shared frailty 0.037* P.H. test 0.182 Exponentiated coefficients* p < 0.05, ** p < 0.01, *** p < 0.001
p-value
(0.131) (0.807) (0.108) (0.001)*** (0.003)** (0.000)*** (0.952) (0.219) (0.079) (0.098) (0.991) (0.916) (0.727)
2,588 5,293 367 0.166 0.814
(3) Non-High-Tech HR YES YES YES 0.817 0.958 1.172 0.884 0.918 0.609 1.045 0.757 1.154 0.917 1.129 0.856 0.996
p-value
(0.128) (0.645) (0.069) (0.000)*** (0.055) (0.000)*** (0.717) (0.073) (0.462) (0.423) (0.088) (0.075) (0.923)
4,188 8,606 707 0.215 0.206
Concentration & Krugman-Diversity Interactions (4) (5) (6) Full Sample High-Tech Non-High-Tech HR YES YES YES 0.791 0.948 1.189 0.868 0.885 0.524 1.052 0.903 1.227 0.863 1.081 0.850 0.979 1.160 1.284 1.137 1.084 1.128 0.912
p-value
(0.041)* (0.459) (0.023)* (0.000)*** (0.001)** (0.000)*** (0.612) (0.358) (0.105) (0.116) (0.198) (0.035)* (0.579) (0.245) (0.038)* (0.185) (0.425) (0.331) (0.461) 6,776 13,899 1,074 0.043* 0.195
HR YES YES YES 0.690 0.965 1.274 0.827 0.818 0.433 0.993 1.061 1.353 0.770 1.011 0.929 0.931 1.054 1.196 1.037 1.252 1.447 0.935
p-value
(0.111) (0.758) (0.110) (0.001)*** (0.005)** (0.000)*** (0.967) (0.725) (0.071) (0.111) (0.909) (0.582) (0.289) (0.850) (0.303) (0.794) (0.235) (0.046)* (0.679) 2,588 5,293 367 0.267 0.768
HR YES YES YES 0.813 0.960 1.172 0.883 0.916 0.612 1.066 0.752 1.151 0.915 1.147 0.782 1.001 1.163 1.418 1.273 1.046 0.862 0.966
p-value
(0.119) (0.661) (0.070) (0.00)*** (0.049)* (0.00)*** (0.599) (0.070) (0.474) (0.411) (0.062) (0.012)* (0.976) (0.307) (0.046)* (0.049)* (0.711) (0.388) (0.869) 4,188 8,606 707 0.200 0.215
5.1 Control and innovation variables For the unreported industry dummies, the results are largely insignificant within the subsamples, indicating that the division into high-tech and non-high-tech leads to a quite homogeneous classification. Significant results for the unreported cohort dummies can be interpreted such that founding conditions are imprinted and play a critical role for survival also in later years after founding (Singh and Lumsden 1990). A founding team consisting only of women or foreigners reduces survival chances, though this effect is significant only in the overall sample. Experience in the industry proves to be a generally beneficial attribute, while no significant relationship is found between academic education and survival. 16
These results suggest that not general, but rather specific qualifications are of importance (Colombo et al. 2004). Company size is positively associated with survival prospects. However, this effect appears to be driven by the conditions in the high-tech subsample, thereby supporting the idea of a liability of smallness in this technological environment (Aldrich and Auster 1986). A very strong and constant significance is found for companies that entered the market as capital companies. According to Doms et al. (1995) and Tveteras and Eide (2000), this can be explained by the high capital requirements for founding such a firm, which leads to a self-selection of promising and well-endowed start-ups. Finally, the insignificant results for the East-West dummy and population density show that there are no differences in survival across these spatial dimensions beyond effects that can be traced back to other variables. Regarding the three innovation variables, introducing market novelties appears to be a strategy without advantages for start-ups in general. This is in line with Buddelmeyer et al. (2010: 281), who state that radical product innovations “should only be undertaken by companies who are financially secure”. Although we do not find any significant relationships between introducing market novelties and survival at the 5% level, some relationships are detected if significance is defined at the 10% level: first, hightech start-ups introducing global market novelties show decreased survival prospects. We might interpret this finding such that high-tech start-ups are under considerable pressure to innovate and engage in overly complex and risky innovation projects. Second, start-ups in non-high-tech environments introducing national market novelties face better survival prospects. We might interpret this such that in non-high tech environments being innovative is a supportive condition for survival. These findings are in line with previous results obtained by Cefis and Marsili (2011), who also find that non-innovative companies in low-tech industries, as well as innovating companies in high-tech industries, have the highest risk of failure. However, it must be stated that all these interpretations of the innovation variables are somewhat afflicted by uncertainty. Cefis and Marsili (2011) show that the relationship between innovative activity and market exit differs according to the exit route. Hence, it is possible that some effects may not be detectable in our analysis, as the influence on exit due to failure is balanced out by the opposing effect of exit due to M&A.
5.2 Interdependence of externalities and innovation Localization externalities 17
Regarding localization externalities, we find that neither regional concentration nor specialization exert a positive main effect in the baseline models (models 1-3). Accordingly, hypothesis 1a cannot be confirmed. However, the measure for regional concentration undergoes a major change when the interaction terms are included. To be more precise, the coefficient for the main effect of regional concentration now indicates a significant and beneficial impact on survival, while the interaction between national market novelties and concentration indicates a significant and detrimental impact on survival (model 4). These findings underpin the importance of considering company characteristics when examining the impact of agglomeration externalities on new firm survival. In line with previous research showing that localization externalities are mainly relevant in traditional sectors (Renski 2011; Weterings and Marsili 2015), a closer examination reveals that this result is driven by the non-high-tech subsample (model 6). For these non-high-tech start-ups we find a significant and detrimental effect for the interaction of global and national novelties with regional concentration. These findings suggest that not all companies are affected by regional concentration in the same way. That is to say, more innovative start-ups in non-high-tech environments seem to benefit less from localization externalities than their non-innovative counterparts. Moreover, the interaction coefficients for national market novelties are even larger than the overall effect of regional concentration. This means that start-ups with national market novelties not only benefit less, but might even be adversely affected by a dense regional concentration of similar firms. As for regional specialization, we do not find any significant main effect (models 7-9). However, the interaction between national market novelties and regional specialization also reveals a significant and detrimental impact on survival for non-high-tech start-ups (model 9). Taken together, our results support hypothesis H2a and the findings of Pe’er and Keil (2013): non-high-tech start-ups with highly innovative products gain fewer advantages from being located in spatial proximity to similar firms. Accordingly, they rely less on exploiting external sources, but might still experience the drawbacks of increased competition when they are located in close spatial proximity to similar firms. Furthermore, applying two distinct measures – regional concentration and regional specialization – permits a more fine-grained interpretation. Precisely, regional concentration appears to be the more powerful predictor of new firm survival. Essentially, this suggests that the disproportionate presence of an industry in a region might not be sufficient to create localization externalities from which newly founded firms can benefit. Rather, it is necessary to have a disproportionately strong spatial density of firms belonging to the same industry. Accordingly, this finding underpins the critical role of geographical proximity for economic interactions and the exchange of knowledge (Boschma 2005). 18
Regional diversity With regard to regional economic diversity, we do not find a significant overall impact for any of our applied measures, neither for high-tech nor for non-high-tech industries (see models 1-9 and appendix 1). Thus, hypothesis H1b cannot be confirmed. The only significant effect that we find in the context of regional economic diversity is a negative relationship for the interaction between Krugman diversity and national market novelties in high-tech environments. This finding holds regardless of the localization measure included (models 4 and 8). Hence, hypothesis H2b is partly confirmed, but we find evidence only for high-tech firms that introduced market novelties with a national scope. Since hypothesis H2a was not confirmed, meaning that no positive effect of Krugman-Diversity was found, the coefficient of the interaction between Krugman-Diversity and national novelties implies that companies with national novelties not only benefit less from, but are even adversely affected by regional economic diversity. Following the arguments that led to hypothesis H2b, this finding could be interpreted such that a) interindustry spillovers seem to be present and b) the recombination of knowledge from different sectors leads to complex and risky innovation projects. Interestingly, this effect does not apply to global novelties. Apparently, in these cases recombining diverse knowledge leads to innovations, which are indeed radical, but, on average, successful enough not to lower the firm’s survival probability. The previously mentioned idea that high-tech start-ups are under considerable pressure to innovate (Cefis and Marsili 2011) might also provide a possible answer to the question as to why non-high-tech start-ups with market novelties are not affected by the diversity of the regional economy. As these companies are under less pressure to introduce these kinds of radical innovations, they might also be under less pressure to find and exploit diverse sources of knowledge. Furthermore, the results can be interpreted as follows: while regional diversity might well trigger the appearance of novelties (which is not studied here), this does not increase, and in the case of high-tech start-ups might even decrease, the survival chances of the start-ups that introduce these novelties.3 Finally, it might appear somewhat surprising that we find an interaction between regional economic diversity and firm innovativeness only for Krugman-Diversity, while the remaining indicators (Herfindahl-Diversity, Related/Unrelated Variety) are not significant. However, as table 1 reveals, all the indicators feature some unique variance and, hence, do not measure perfectly identical constructs. To be more precise, the Krugman diversity measure assesses regional economic diversity in relation to the 3
We thank an anonymous referee for this idea.
19
industry mix of the whole nation. Consequently, the presence of an industry that is marginal in absolute terms but disproportionate in relation to the overall industry mix can have a considerable impact on this indicator. In contrast, the measures of Herfindahl-Diversity and Related/Unrelated Variety depict the unqualified industry composition in a region. Accordingly, this indicator represents the prevalence of a few dominating industries to a greater extent, while industries with low employment figures do not contribute much variance to the indicator. Finding an effect only for Krugman -Diversity suggests that it might not be the sheer presence of many different large industries that drives the positive effect on survival, but rather that certain industries might be exceptionally relevant for innovative start-ups although their absolute share of the regional economy is only marginal.
6 Limitations and robustness checks Although the presented results largely match the hypotheses, they also possess elements of uncertainty and lead to some unexpected results. The main effects of our indicators for regional economic diversity do not even come close to exhibiting a statistically significant influence (see table 3 and appendix 1). On the one hand, these null effects for regional diversity are somewhat in line with the very few and mixed findings on the relationship between regional diversity and new firm survival (Basile et al. 2016; Howell et al. 2016; Renski 2011). On the other hand, it is also possible that we did not find any effects due to misspecifications in our data. One possible reason for the unexpected results might lie in the operationalization of the diversity measurements. The tests for shared frailty indicate that there is an unobserved correlation among entities belonging to the same labor market region in some models. Accordingly this might indicate that not all regional effects are adequately integrated into the model. However, the significant effect of shared frailty is only present in the full sample model and vanishes when the analysis is conducted separately for the subsamples. Furthermore, we included different measures of regional economic diversity and our specifications largely follow previous studies (Basile et al. 2016; Howell et al. 2016; Renski 2011), which detected some significant relationships for these measures. Another possible reason for an empirical misspecification could be that the regional scope applied is still too small. However, the regional level was chosen carefully and the presence of shared frailty within these spatial entities undermines this argument. Another concern might be that innovations leading to market novelties are such a rare occurrence that 20
our study fails to detect potentially meaningful interaction effects due to insufficient power. However, the stratification criteria of the underlying sample lead to an over-representation of highly innovative start-ups. To be more specific, of the total number of 6,776 start-ups in our sample, 641 (236 HT/ 405 Non-HT) introduced a regional novelty, 798 (442 HT/ 356 Non-HT) introduced a national novelty and 525 (348 HT/ 177 Non-HT) introduced a global novelty. Given these frequencies, we argue that insufficient power should not be a major issue in our analyses. Nevertheless, we formed alternative innovation measures to boost the statistical power. Specifically, we created one measure grouping all three types of market novelties and an additional measure grouping national and global market novelties. The results of these additional models are available upon request. In line with our previous results, we find for both specifications a) no significant main effect, b) a significant interaction with the concentration measure and c) no significant interaction with our diversity measures. Grouping the different innovations caused the significant interaction between national novelties and diversity in hightech sectors (possessing sufficient power as based on 442 cases) to vanish. We conclude from this finding that the moderating effect of innovativeness on Jacobs externalities does indeed differ according to the scope of the market novelty, which is an argument in favor of a more nuanced analysis. A further source of uncertainty lies in the ability to distinguish between different ways of exiting the market. This is critical, as previous empirical findings revealed that the determinants of exit vary according to the mode of exit. It is possible that potentially hypothesis-conforming results for regional economic diversity may not be detectable, because the influence of regional characteristics on exit due to failure is balanced out by the opposing effect on exit due to M&A. However, Coad (2014) shows that, in the vast majority of cases, exit does indeed reflect the company’s lack of success. Furthermore, if the inability to distinguish between exit routes was the main reason for the unpredicted results, it is hard to explain why other parameters behave as expected. Either our variables for regional economic diversity are the only ones showing opposing effects on different modes of exit – which appears unlikely and would be hard to reason – or the potential influence was very weak anyway, so that even small biases cause it to vanish. Other possibilities for unpredicted results have to be sought in the underlying data. First, the study is based on a new database, which differs from previous databases in such crucial areas as sectoral composition, definition of company entry/death and minimum employment thresholds. Accordingly, transferring results from existing studies is necessarily associated with uncertainty. Finally, with hazard rates typically reaching a peak at around two years (van Praag 2003, Brüderl et al. 2007), our observation period of four years still covers the most critical phase after founding. However, 21
empirical results suggest that in some industries hazard rates reach their maximum much later - after around seven years (Agarwal and Audretsch 2001). With respect to the innovation variables, this study overcame some drawbacks of previous investigations. However, the study design does not deliver any information about start-ups engaged in innovative activities that do not lead to market novelties, such as process and incremental product innovations. Further potential biases in the model might be associated with non-random panel attrition, which results in only start-ups with certain characteristics remaining in the panel. To test for this bias, an additional Cox regression is run, whereby the failure event now constitutes surviving start-ups which exit from the panel before the end of the analysis4. The results of this analysis are reported in appendix 2 and show that most of the variables are not significantly related to the probability of panel attrition. However, significant results are found for a few variables. To address this potential bias, the original survival models (models 1-6) are repeated in a parametric accelerated failure time (AFT) model with log-logistic distribution (Hutton and Monaghan 2002). As AFT models assign a prominent role to survival duration, they are less vulnerable to data loss by censoring (Cader and Leatherman 2011). The results are reported in appendix 3 and reveal no serious differences in algebraic signs and significances, which would challenge the interpretation of the results of the Cox model. Accordingly, the results can be considered robust against model specification and potential sample selection bias. Finally, when analyzing the influence of geographic characteristics on new firm survival, problems of geographic self-selection might arise. In this context, self-selection would mean that the objects select themselves into a region with certain characteristics, leading to a generally biased result. Renski (2011) names location choice and the founder’s experience as potential sources of geographic self-selection. However, empirical evidence shows that start-ups are usually not subject to complex location decisions (Nerlinger 1999; Mossig 2000). Regarding the founder’s experience, the present study is able to control for the founder’s industry experience and shows that the applied study design, with control variables accounting for initial firm and founder heterogeneity, generally makes the presence of a heavy bias due to geographic self-selection unlikely.
4
A violation of the proportional hazards assumption required a stratification of the model according to founding cohorts.
22
7 Conclusion This paper provides evidence that the effect of agglomeration externalities on survival is, indeed, moderated by the innovative behavior of start-ups. It becomes clear that localization externalities are prevalent in non-high-tech environments. Furthermore, the most innovative start-ups are less positively or even negatively affected by localization externalities, which means that especially the less innovative companies appear to benefit from being located in spatial proximity to similar firms. As for regional diversity, in most cases no significant relationship is found between regional economic diversity and new firm survival. However, high-tech start-ups with national market novelties are negatively affected by a diverse regional economic structure. This effect can be interpreted as indirect evidence of the presence of inter-industry spillovers, leading to risky innovations. Finally, these results not only contribute to scientific knowledge, but also bear some important practical implications for policymakers. The control variables included show that basic demographic features such as gender or nationality are likely to determine new firm survival. These results demonstrate structural barriers which systematically disadvantage certain social groups. Eliminating these handicaps could not only lead to a higher number of surviving start-ups, but is ultimately a matter of social justice. Furthermore, the study shows that introducing market novelties is not beneficial for start-ups and that recombining diverse knowledge does not always pay off. Accordingly, providing support to firms that are engaged in such innovation projects could constitute another important approach for policymakers. This could take the form of financial support, such as cheap loans or funding, or could be done by providing information and consultation to better assess the risks of highly innovative projects.
23
References Agarwal, R., & Audretsch, D. B. (2001). Does entry size matter? The impact of the life cycle and technology on firm survival. Journal of Industrial Economics, 49, 21-43. doi:10.1111/1467-6451.00136 Aldrich, H.E., & Auster, E.R. (1986). Even dwarfs started small: Liabilities of age and size and their strategic implications. In B.M. Staw & L.L. Cummings (Eds.), Research in Organizational Behavior, vol. 8 (pp. 165-198). Greenwich: JAI Press. Armington, C., & Acs, Z.J. (2002). The Determinants of Regional Variation in New Firm Formation. Regional Studies, 36, 33-45. doi; 10.1080/00343400120099843 Arrow, K. J. (1962). Economic welfare and allocation of resources for invention. In: National Bureau Committee for Economic Research (Ed.), The rate and direction of inventive activity: Economic and social factors, Princeton: Princeton University Press Audretsch, D.B. (1991). New Firm Survival and the Technological Regime. The Review of Economics and Statistics, 68, 441450. doi: 10.2307/2109568 Basile, R., Pittiglio, R., & Reganati F. (2016). Do agglomeration externalities affect firm survival?. Regional Studies, 51, 548562. doi: 10.1080/00343404.2015.1114175 BBSR (2017). Laufende Raumbeobachtung – Raumabgrenzungen – Arbeitsmarktregionen. http://www.bbsr.bund.de/BBSR/DE/Raumbeobachtung/Raumabgrenzungen/AMR/Arbeitsmarktregionen.html?nn=443270. Accessed 19 May 2017. Boschma, R. (2005). Proximity and Innovation: A Critical Assessment. Regional Studies, 39, 61-74. doi: http://dx.doi.org/10.1080/0034340052000320887 Boschma, R., & Iammarino, S. (2009). Related Variety, Trade Linkages, and Regional Growth in Italy. Economic Geography, 85, 289-311. doi: 10.1111/j.1944-8287.2009.01034.x Boyer, T., & Blazy, R. (2014). Born to be alive? The survival of innovative and non-innovative French micro-start-ups. Small Business Economics, 42, 669-683. doi: 10.1007/s11187-013-9522-8 Brixy U., & Grotz, R. (2007). Regional patterns and determinants of birth and survival of new firms in Western Germany. Entrepreneurship & Regional Development – An International Journal, 19, 293-312. doi: 10.1080/08985620701275510 Brüderl, J., & Schüssler, R. (1990). Organizational Mortality: The liabilities of newness and adolescence. Administrative Science Quarterly, 35, 530-547. doi: 10.2307/2393316 Brüderl, J., Preisendörfer, P., & Ziegler, R. (1992). Survival Chances of Newly Founded Business Organizations. American Sociological Review, 57, 227-242. doi: 10.2307/2096207 Brüderl, J., Preisendörfer, P., & Ziegler, R. (2007). Der Erfolg neu gegründeter Betriebe. Eine empirische Studie zu den Chancen und Risiken von Unternehmensgründungen, 3rd edition. Berlin: Duncker & Humblot. Buddelmeyer, H., Jensen, P.H., & Webster, E. (2010). Innovation and the determinants of company survival. Oxford Economic Papers, 62, 261-285. doi: 10.1093/oep/gpp012 Cader, H.A., & Leatherman, J.C. (2011). Small business survival and sample selection bias. Small Business Economics, 37, 155-165. doi: 10.1007/s11187-009-9240-4 Cainelli, G., Montresor, S., & Marzetti, G.V. (2014). Spatial agglomeration and firm exit: a spatial dynamic analysis for Italian provinces. Small Business Economics, 43, 213-228. doi: 10.1007/s11187-013-9532-6 Cefis, E., & Marsili, O. (2006). Innovation Premium and the Survival of Entrepreneurial Firms in the Netherlands. International Studies in Entrepreneurship, 12, 183-198. doi: 10.1007/0-387-32314-7_9
24
Cefis, E., & Marsili, O. (2011). Born to flip. Exit decisions of entrepreneurial firms in high-tech and low-tech industries. Journal of Evolutionary Economics, 21, 473-498. doi: 0.1007/s00191-010-0210-4 Coad, A. (2014). Death is not a success: reflections on business exit. International Small Business Journal, 32, 721-732. doi: 10.1177/0266242612475104 Colombo, M., Delmastro, M., & Grilli, L. (2004). Entrepreneurs’ human capital and the start-up size of new technology-based firms. International Journal of Industrial Organization, 22, 1183-1211. doi: 10.1016/j.ijindorg.2004.06.006 Cox, D.R. (1972). Regression models and life-tables (with discussion). Journal of the Royal Statistical Society, B34, 187-220. doi: 10.1007/978-1-4612-4380-9_37 Dauth, W. (2013). Agglomeration and regional employment dynamics. Papers in Regional Science, 92, 419-435. doi: 10.1111/j.1435-5957.2012.00447.x Désiage, L., Duhautois, R., & Redor, D. (2010). Do Public Subsidies Have an Impact on New Firm Survival? An Empirical Study with French Data. Working Paper 2010-04, TEPP - Institute for Labor Studies and Public Policies. doi: 10.2139/ssrn.1640560 Doms, M., Dunne, T., & Roberts, M.J. (1995). The role of technology use in the survival and growth of manufacturing plants. International Journal of Industrial Organization, 13, 523-542. doi: 10.1016/0167-7187(95)00503-X Eckey, H.-F., Kosfeld, R., & Türck, M. (2006). Abgrenzung deutscher Arbeitsmarktregionen. Raumforschung und Raumordnung, 64, 299-309. doi: 10.1007/BF03183178 Falck, O. (2007). Survival chances of new businesses: do regional conditions matter?. Applied Economics, 39, 2039-2048. doi: 10.1080/00036840600749615 Federal Statistical Office (2008). Klassifikation der Wirtschaftszweige – Mit Erläuterungen. Wiesbaden: Statistisches Bundesamt. Fichman, M., & Levinthal, D.A. (1991). Honeymoons and the liability of adolescence: a new perspective on duration dependence in social and organizational relationships. Academy of Management Review, 16, 442-46. doi: 10.5465/AMR.1991.4278962 Folta, T.B., Cooper, A.C., & Baik, Y.S. (2006). Geographic cluster size and firm performance. Journal of Business Venturing, 21, 217-242. doi: 10.1016/j.jbusvent.2005.04.005 Fontana, R., & Nesta, L. (2009). Product Innovation and Survival in a High-Tech Industry. Review of Industrial Organization, 34, 287-306. doi: 10.1007/s11151-009-9210-7 Frenken, K., Van Oort, F., & Verburg, T. (2007). Related Variety, Unrelated Variety and Regional Economic Growth. Regional Studies, 41, 685-697. doi; 10.1080/00343400601120296 Fritsch, M., & Falck, O. (2007). New Business Formation by Industry over Space and Time: A Multidimensional Analysis. Regional Studies, 41, 157-172. doi: 10.1080/00343400600928301 Fryges, H., Gottschalk, S., & Kohn, K. (2009). The KfW/ZEW Start-up Panel: Design and Research Potential. ZEW Research Discussion Paper, No. 09-053, Mannheim: ZEW. doi: 10.2139/ssrn.1494326 Fujita, M., & Thisse, J.-F. (1996). Economics of agglomeration. Journal of the Japanese and International Economies, 10, 339–378. Geroski, P.A., Mata, J., & Portugal, P. (2010). Founding conditions and the survival of new firms. Strategic Management Journal, 31, 510-529. doi: 10.1002/smj.823 Glaeser, E., Kallal, H., Scheinkman, J., & Schleifer, A. (1992). Growth in Cities. Journal of Political Economy, 100, 11261152. doi: 10.3386/w3787
25
Grambsch, P.M. & Therneau, T.M. (1994). Proportional hazards tests and diagnostics based on weighted residuals. Biometrika, 81, 515-526. doi: 10.2307/2337547 Gruhl, A., Schmucker, A,. & Seth, S. (2012). The Establishment History Panel 1975-2010. FDZ Datenreport 04/2012, Research Data Centre of the German Federal Employment Agency at the Institute for Employment Research. Harhoff, D., Stahl, K., & Woywode, M. (1998). Legal form, growth and exit of West German firms. Journal of Industrial Economics, 46, 453-488. doi: 10.1111/1467-6451.00083 Helmers, C., & Rogers, M. (2010). Innovation and the Survival of New Firms in the UK. Review of Industrial Organization, 36, 227-248. doi; 10.1007/s11151-010-9247-7 Henderson, J.V., Kuncoro, A., & Turner, M. (1995). Industrial development in cities. Journal of Political Economy, 103, 1067–1090. doi: 10.3386/w4178 Howell A., He, C., Yang, R., & Fan, C. (2016). Agglomeration, (un)-related variety and new firm survival in China: Do local subsidies matter?. Papers in Regional Science, doi: 10.1111/pirs.12269 Hutton, J.L., & Monaghan, P.F. (2002). Choice of Parametric Accelerated Life and Proportional Hazards Models for Survival Data: Asymptotic Results. Lifetime Data Analysis, 8, 375- 393. doi: 10.1023/A:1020570922072 Hyytinen, A., Pajarinen, M. & Rouvinen, P. (2015). Does innovativeness reduce startup survival rates?. Journal of Business Venturing, 30, 564-581. doi: 10.1016/j.jbusvent.2014.10.001 Jacobs, J. (1969). The economy of cities. New York: Vintage. Jensen, P.H., Webster, E., & Buddelmeyer, H. (2008). Innovation, Technological Conditions and New Firm Survival. The Economic Record, 84, 434-448. doi: 10.2139/ssrn.946827 Knaup, A.E. (2005). Survival and longevity in the business employment dynamics data. Monthly Labor Review, 117, 2-8. Legler, H., & Frietsch, R. (2006). Neuabgrenzung der Wissenswirtschaft – forschungsintensive Industrien und wissensintensive Dienstleistungen (NIW/ISI-Listen 2006). Studien zum deutschen Innovationssystem Nr.22-2007, Hannover: Niedersächsisches Institut für Wirtschaftsforschung. Marshall, A. (1920). Principles of economics. London: Macmillan. Mata, J., & Portugal, P. (1994). Life duration of new firms. Journal of Industrial Economics, 42, 227-246. doi: 10.2307/2950567 McCann, B.T., & Folta, T.B. (2008). Location Matters: Where We Have Been and Where We Might Go in Agglomeration Research. Journal of Management, 34, 532-565. doi: 10.1177/0149206308316057 Mossig, I. (2000). Räumliche Konzentration der Verpackungsmaschinenbau-Industrie in Westdeutschland. Eine Analyse des Gründungsgeschehens. Münster/Hamburg/London: Lit-Verlag. Neffke, F., & Henning, M. (2008). Revealed Relatedness: Mapping Industry Space. DRUID Working Papers 08-18. Neffke, F.M.H., Henning, M., & Boschma, R. (2012). The impact of aging and technological relatedness on agglomeration externalities: a survival analysis. Journal of Economic Geography, 12, 485-517. doi: 10.1093/jeg/lbr001 Nerlinger, E.A. (1999). Innovative Unternehmensgründungen in Deutschland: ein Überblick über aktuelle Forschungsergebnisse und Trends. In D. Bögenhold (Ed.) Eine neue Gründerzeit? Die Wiederentdeckung kleiner Unternehmen in Theorie und Praxis (pp. 111-136). Amsterdam: G & B Verl. Fakultas. Nooteboom, B., Van Haverbeke, W., Duysters, G., Gilsing, V., & Van den Oord, A. (2007). Optimal cognitive distance and absorptive capacity. Research Policy, 36, 1016-1034. doi: http://dx.doi.org/10.1016/j.respol.2007.04.003
26
OECD (2005). Oslo Manual – Guidelines for Collecting and Interpreting Innovation Data, 3rd edition. Paris: Organisation for Economic Co-Operation and Development and Statistical Office of the European Communities. doi; 10.1787/9789264013100-en Pe’er, A., & Keil, T. (2013). Are all startups affected similarly by clusters? Agglomeration, competition, firm heterogeneity, and survival. Journal of Business Venturing, 28, 354-372. doi: 10.1016/j.jbusvent.2012.03.004 Pe’er, A., Vertinsky, I., & Keil, T. (2016). Growth and survival: The moderating effects of local agglomeration and local market structure. Strategic Management Journal, 37, 541-564. doi: 10.1002/smj.2331 Prevezer, M. (1997). The dynamics of industrial clustering in biotechnology. Small Business Economics, 9, 255-271. doi: 10.1023/A:1017995006003 Puig, F., González-Loureiro, M., & Ghauri, P.N. (2014). Internationalisation for Survival: The Case of New Ventures. Management International Review, 54, 653-673. doi; 10.1007/s11575-014-0209-4 Renski, H. (2011). External economies of localization, urbanization and industrial diversity and new firm survival. Papers in Regional Science, 90, 473-502. doi: 10.1111/j.1435-5957.2010.00325.x Renski, H. (2015). Externalities or Experience? Localization Economies and Start-up Business Survival. Growth and Change, 46, 458-480. doi: 10.1111/grow.12099 Rigby, D.L. (2015). Technological Relatedness and Knowledge Space: Entry and Exit of US Cities from Patent Classes. Regional Studies, 49, 1922-1937. doi: http://dx.doi.org/10.1080/00343404.2013.854878 Romer, P. M. (1986). Increasing returns and long-run growth. Journal of Political Economy, 94, 1002–1037. Rosenbusch, N., Brinckmann, J., & Bausch, A. (2011). Is innovation always beneficial?: a meta-analysis of the relationship between innovation and performance in SMEs. Journal of Business Venturing, 26, 441-457. doi: 10.1016/j.jbusvent.2009.12.002 Schmalensee, R. (1978). Entry deterrence in the ready-to-eat breakfast cereal industry. Bell Journal of Economics, 9, 305327. doi; 10.2307/3003584 Schutjens, V.A.J.M., & Wever, E. (2000). Determinants of new firm success. Papers in Regional Science, 79, 135-159. doi: 10.1007/s101100050040 Scitovsky, T. (1954). Two concepts of external economies. Journal of Political Economy, 62, 143–151. Singh, J.V., & Lumsden, C.J. (1990). Theory and Research in Organizational Ecology. Annual Review of Sociology, 16, 161195. doi: 10.1146/annurev.so.16.080190.001113 Stearns, T.M., Carter, N.M., Reynolds, P.D., & Williams, M.L. (1995). New firm survival: Industry, strategy, and location. Journal of Business Venturing, 10, 23-42. doi: 10.1016/0883-9026(94)00016-N Sternberg, R., & Litzenberger, T. (2004). Regional clusters in Germany: Their geography and their relevance for entrepreneurial activities. European Planning Studies, 12, 767-791. doi: 10.1080/0965431042000251855 Strotman, H. (2007). Entrepreneurial Survival. Small Business Economics, 28, 87-104. doi: 10.1007/s11187-005-8859-z Tödtling, F., Lehner, P., & Kaufmann A. (2009). Do different types of innovation rely on specific kinds of knowledge interactions?. Technovation, 29, 59-71. doi: 10.1016/j.technovation.2008.05.002 Tveteras, R., & Eide, G.E. (2000). Survival of new plants in different industry environments in Norwegian manufacturing: a semi-proportional Cox model approach. Small Business Economics, 14, 65-82. doi: 10.1023/A:1008189411442 Van Praag, M. (2003). Business Survival and Success of Young Small Business Owners. Small Business Economics, 21, 117. doi: 10.2139/ssrn.289202
27
Wennberg, K. & Lindqvist, G. (2010). The effect of clusters on the survival and performance of new firms. Small Business Economics, 34, 221-241. doi: 10.1007/s11187-008-9123-0 Weterings, A., & Marsili, O. (2015). Spatial Concentration of Industries and New Firm Exits: Does this Relationship Differ between Exits by Closure and by M&A?. Regional Studies, 49, 44-58. doi: 10.1080/00343404.2012.726708 Wyrwich, M. (2013). Can socioeconomic heritage produce a lost generation with regard to entrepreneurship?. Journal of Business Venturing, 28, 667-682. doi: 10.1016/j.jbusvent.2012.09.00
28
Appendix Appendix 1: Results of Cox regression with Specialization, Herfindahl-Diversity and Related/Unrelated Variety Interactions.
Model Condition
Specialization & Krugman-Diversity Interactions (1) (2) (3) Full Sample High-Tech Non-High-Tech
HR Cohort controls YES Sectorial controls YES KfW funding control YES German_Involved 0.792 Quali_Academic 0.949 Women_only 1.192 Experience_Log 0.868 Employ_Log 0.885 Legal_Form 0.524 Regional_Novelty 1.041 National_Novelty 0.898 Global_Novelty 1.231 East_West 0.867 Pop_Density 1.017 Concentration Specialization 0.907 Krugman-Diversity 0.970 Herfindahl-Diversity Rel.Var. Unrel.Var. Reg_Nov#Conc. Nat_Nov#Conc. Glob_Nov#Conc. Reg_Nov# Spec. 1.164 Nat_Nov# Spec. 1.204 Glob_Nov# Spec. 1.143 Reg_Nov#Krugm.-Div. 1.104 Nat_Nov#Krugm.-Div. 1.164 Glob_Nov#Krugm.-Div. 0.934 Reg_Nov#Herf.-Div. Nat_Nov# Herf.-Div. Glob_Nov# Herf.-Div. Reg_Nov# Rel.Var. Nat_Nov# Rel.Var. Glob_Nov# Rel.Var. Reg_Nov# Unrel.Var. Nat_Nov# Unrel.Var. Glob_Nov# Unrel.Var. N subjects N annual spells N exits Shared frailty P.H. test Exponentiated coefficients * p < 0.05, ** p < 0.01, *** p < 0.001
p-value
p-value
(0.043)* (0.474) (0.021)* (0.000)*** (0.001)** (0.000)*** (0.688) (0.333) (0.097) (0.133) (0.712)
HR YES YES YES 0.716 0.958 1.274 0.826 0.817 0.433 0.927 1.061 1.346 0.773 0.995
(0.245) (0.438)
0.918 0.922
(0.218) (0.040)* (0.360) (0.316) (0.205) (0.583)
0.454 1.212 1.084 1.302 1.506 0.945
p-value
(0.152) (0.715) (0.110) (0.001)*** (0.004)** (0.000)*** (0.718) (0.722) (0.073) (0.116) (0.941)
HR YES YES YES 0.808 0.965 1.175 0.883 0.919 0.611 1.045 0.738 1.154 0.920 1.029
(0.619) (0.220)
0.897 0.997
(0.260) (0.945)
(0.508) (0.285) (0.683) (0.143) (0.024)* (0.721)
1.172 1.238 1.172 1.059 0.894 1.007
(0.109) (0.697) (0.067) (0.00)*** (0.057) (0.00)*** (0.716) (0.055) (0.464) (0.442) (0.565)
Concentration & Herfindahl-Diversity Interactions (4) (5) (6) Full Sample High-Tech Non-High-Tech HR YES YES YES 0.788 0.947 1.186 0.868 0.884 0.526 1.056 0.926 1.230 0.869 1.085 0.843
p-value
p-value
(0.038)* (0.455) (0.024)* (0.000)*** (0.001)** (0.000)*** (0.584) (0.482) (0.101) (0.137) (0.172) (0.028)*
HR YES YES YES 0.697 0.968 1.256 0.830 0.818 0.437 0.990 1.144 1.362 0.765 1.026 0.888
1.037
p-value
(0.121) (0.777) (0.134) (0.001)*** (0.005)** (0.000)*** (0.957) (0.393) (0.064) (0.104) (0.794) (0.373)
HR YES YES YES 0.813 0.961 1.170 0.882 0.916 0.611 1.067 0.776 1.148 0.922 1.148 0.785
(0.242)
1.048
(0.323)
1.026
(0.517)
(0.119) (0.665) (0.073) (0.000)*** (0.048)* (0.000)*** (0.595) (0.107) (0.485) (0.453) (0.044)* (0.013)*
1.179 1.333 1.129
(0.179) (0.014)* (0.210)
1.168 1.291 1.032
(0.543) (0.127) (0.816)
1.166 1.449 1.265
(0.287) (0.034)* (0.062)
0.931 1.027 1.228
(0.564) (0.893) (0.307)
0.785 0.778 1.420
(0.529) (0.419) (0.168)
0.963 1.571 0.963
(0.769) (0.115) (0.919)
Concentration & Related/Unrelated Variety Interactions (7) (8) (9) Full Sample High-Tech Non-High-Tech HR YES YES YES 0.793 0.955 1.186 0.868 0.884 0.525 1.055 0.801 1.230 0.863 1.093 0.849
p-value
1.030 0.934 1.163 1.308 1.138
2,588 5,293 367 0.257 0.909
p-value
(0.133) (0.812) (0.140) (0.01)*** (0.005)** (0.000)*** (0.983) (0.425) (0.073) (0.107) (0.877) (0.711)
HR YES YES YES 0.812 0.966 1.171 0.883 0.915 0.611 1.069 0.765 1.186 0.910 1.158 0.780
(0.466) (0.104) (0.236) (0.030)* (0.168)
1.068 0.875 1.116 1.187 1.010
(0.364) (0.053) (0.685) (0.336) (0.944)
1.024 0.962 1.157 1.502 1.267
(0.613) (0.437) (0.329) (0.026)* (0.049)*
0.938 0.876 0.926 1.125 1.370 0.980
(0.722) (0.490) (0.687) (0.552) (0.114) (0.910) 2,588 5,293 367 0.189 0.484
0.965 0.934 1.153 1.066 0.817 0.816
(0.118) (0.709) (0.072) (0.00)*** (0.045)* (0.00)*** (0.584) (0.088) (0.384) (0.393) (0.034)* (0.012)*
(0.223) (0.046)* (0.553) (0.635) (0.498) (0.974)
0.960 0.940 1.067 1.075 1.060 0.849 6,776 13,899 1,074 0.028* 0.257
p-value
(0.044)* (0.520) (0.024)* (0.000)*** (0.001)** (0.000)*** (0.588) (0.096) (0.101) (0.121) (0.132) (0.035)*
HR YES YES YES 0.700 0.973 1.252 0.829 0.820 0.431 1.004 1.135 1.349 0.763 1.015 0.951
4,188 8,606 707 0.257 0.909
6,776 13,899 1,074 0.046* 0.127
2,588 5,293 367 0.231 0.676
4,188 8,606 707 0.200 0.215
(0.691) (0.627) (0.644) (0.520) (0.671) (0.250) 6,776 13,899 1,074 0.052 0.101
(0.780) (0.699) (0.534) (0.642) (0.288) (0.405) 4,188 8,606 707 0.211 0.361
Appendix 2: Results of test for non-random panel attrition.
Results of Cox regression for full sample with panel attrition as failure event Sectorial controls KfW funding control German_Involved Quali_Academic Women_only Experience_Log Employ_Log Legal_Form East_West Pop_Density Concentration Krugman-Diversity Regional_Novelty National_Novelty Global_Novelty Number of Subjects Number of annual spells Numer of exits P.H.-test Stratified by cohorts Exponentiated coefficients * p < 0.05, ** p < 0.01, *** p < 0.001
HR YES YES 0.794 0.805 1.015 0.913 0.981 1.098 0.963 0.989 1.016 1.047 0.970 1.118 0.893
p-value (0.010)* (0.000)*** (0.810) (0.000)*** (0.470) (0.101) (0.557) (0.733) (0.584) (0.062) (0.723) (0.131) (0.268) 4,501 10,813 1,919 0.884
Appendix3: Results of robustness check in parametric model setting.
Model Condition
(1) Full Sample
TR Cohort controls YES Sectorial controls YES KfW funding control YES German_Involved 1.226 Quali_Academic 1.040 Women_only 0.894 Experience_Log 1.107 Employ_Log 1.084 Legal_Form 1.541 Regional_Novelty 0.967 National_Novelty 1.057 Global_Novelty 0.883 East_West 1.087 Pop_Density 0.952 Concentration 1.058 Krugman-Diversity 1.017 Reg_Nov#Conc. Nat_Nov#Conc. Glob_Nov#Conc. Reg_Nov#Krugm.-Div. Nat_Nov#Krugm.-Div. Glob_Nov#Krugm.-Div. N subjects N annual spells N exits Shared frailty Exponentiated coefficients * p < 0.05, ** p < 0.01, *** p < 0.001
p-value
(0.014)* (0.435) (0.037)* (0.000)*** (0.002)** (0.000)*** (0.645) (0.457) (0.160) (0.218) (0.247) (0.205) (0.512)
6,776 13,899 1,074 0.017*
Baseline Model (2) High-Tech TR YES YES YES 1.311 1.013 0.850 1.134 1.148 1.772 1.041 0.892 0.836 1.180 0.997 1.009 1.019
p-value
(0.095) (0.876) (0.126) (0.001)** (0.004)** (0.000)*** (0.747) (0.274) (0.125) (0.145) (0.964) (0.883) (0.644)
2,588 5,293 367 0.127
(3) Non-High-Tech TR YES YES YES 1.195 1.035 0.902 1.096 1.056 1.386 0.942 1.242 0.912 1.038 0.914 1.116 1.010
p-value
(0.067) (0.596) (0.103) (0.000)*** (0.078) (0.000)*** (0.505) (0.048)* (0.513) (0.642) (0.087) (0.081) (0.753)
4,188 8,606 707 0.137
Concentration & Krugman-Diversity Interactions (4) (5) (6) Full Sample High-Tech Non-High-Tech TR YES YES YES 1.224 1.038 0.894 1.108 1.085 1.543 0.959 1.087 0.882 1.087 0.944 1.117 1.021 0.919 0.817 0.930 0.944 0.931 1.075
p-value
(0.015)* (0.462) (0.039)* (0.000)*** (0.002)** (0.000)*** (0.570) (0.282) (0.164) (0.219) (0.189) (0.044)* (0.462) (0.360) (0.024)* (0.288) (0.435) (0.411) (0.413) 6,776 13,899 1,074 0.021*
TR YES YES YES 1.331 1.017 0.847 1.140 1.146 1.779 1.041 0.972 0.829 1.174 0.995 1.037 1.055 0.975 0.848 1.011 0.866 0.780 1.066
p-value
(0.078) (0.838) (0.119) (0.001)*** (0.004)** (0.000)*** (0.747) (0.806) (0.111) (0.154) (0.945) (0.695) (0.263) (0.892) (0.164) (0.910) (0.263) (0.049)* (0.567) 2,588 5,293 367 0.189
TR YES YES YES 1.198 1.033 0.902 1.096 1.059 1.375 0.928 1.249 0.918 1.040 0.904 1.188 1.005 0.909 0.772 0.854 0.964 1.145 1.014
p-value
(0.062) (0.620) (0.103) (0.00)*** (0.064) (0.00)*** (0.406) (0.045)* (0.546) (0.625) (0.053) (0.014)* (0.890) (0.379) (0.043)* (0.048)* (0.695) (0.271) (0.927) 4,188 8,606 707 .127