Local knowledge spillovers and the effects of related and unrelated variety on the novelty of innovation

ABSTRACT Knowledge spillovers have been identified as a factor affecting the unequal distribution of innovation in space. In this paper we aim to understand how territorial factors shape the novelty degree of innovations. Thus, we perform an empirical analysis that relates territorial factors to innovative performance at the firm level. Our results show that local knowledge spillovers from research and development expenditures are positively associated with upper-level innovation, while local knowledge spillovers from total innovation expenditures are not related to the degree of novelty of innovation. Furthermore, the impacts on innovation are also moderated by related and unrelated varieties since firms in regions with higher regional-related variety are less likely to generate upper-level innovation.


INTRODUCTION
Firms' innovation performances are related not only to differences in the inputs that they use but also to the specific structural conditions of the places in which they are localized. The heterogeneous spatial distribution of innovation inputs affects the level of localized knowledge externalities and eventually influences innovative outputs. The association between territorial dynamics and innovation has been increasingly studied in the literature, and several analyses have found empirical evidence on the impact of local drivers on innovation in regions (Antonelli & Colombelli, 2017;Crescenzi & Jaax, 2016;Rodríguez-Pose & Villarreal Peralta, 2015;Tavassoli & Carbonara, 2014). However, previous studies have paid little attention to how territorial factors impact innovation at the firm level, and especially to the impacts of these factors on the novelty degree of innovation in developing countries (Antonelli & Colombelli, 2017;Barbosa et al., 2014;Plechero & Chaminade, 2016;Rodríguez-Pose & Villarreal Peralta, 2015). Using a rather rare dataset, this paper contributes to filling this gap by presenting new empirical evidence on why territorial factors matter differently for the degree of novelty of innovation, applying to an emerging country, such as Brazil.
Studies of the novelty degrees of innovation at the firm level allow the identification of the different drivers of radical (or upper-level) innovation and of incremental (or lower-level) innovation. Firms that implement upper-level innovation can benefit from stronger competitive positions based on the creation of highly differentiated assets and capabilities. Upper-level innovation is associated with higher risks and requires higher innovative expenditures in comparison to lower-level innovation. In this way, we can assume that different degrees of innovation are driven by distinct innovative efforts (Audretsch & Belitski, 2020;Barbosa et al., 2014;Fitjar & Rodríguez-Pose, 2013;Mascarini et al., 2022;Schoenmakers & Duysters, 2010).
The purpose of this paper is to understand how territorial factors shape the degree of novelty of innovation. This latter is defined following the innovation survey methodology and it is implemented using the categories new-to-the-firm, new-to-the-industry and new-to-theword. Regarding territorial factors, we focus on the impact of knowledge spillovers and related and unrelated factors on the capacity to introduce innovations belonging to one of the abovementioned categories.
Our theoretical background is based on the relationship between upper-level innovations and complex knowledge. Accordingly, we expected that local knowledge spillovers that foster upper-level innovations come from more complex knowledge sources, such as research and development (R&D) expenditures. On the other hand, to foster lower-level innovations, local knowledge spillovers from less complex sources, such as expenses on the external acquisition of embodied technology, would be sufficient. Moreover, we study the effect of related and unrelated variety, drawing on the established tenet according to which related variety is more associated with incremental innovation, since new knowledge is built on established cognitive structures across related knowledge domains. Conversely, unrelated variety is associated with upper-level innovation since it offers the building blocks for technological breakthroughs, stemming from combination across unrelated knowledge.
This paper adds to the literature in several aspects. First, while the literature has concentrated on the influence of territorial factors on regional innovation, there is scarce consideration of how they affect firm-level innovation. Second, the literature on the novelty degree of innovation is gaining momentum. Most empirical studies focus on patent-level analyses, while there is little evidence about firm-level dynamics and the interplay with regional knowledge bases. Third, there is a lack of literature on this subject in developing countries. Our focus on Brazil can help shed new light on innovation dynamics in these contexts. In sum, we are using a unique dataset and applying an empirical strategy, although standard, that allows us to comprehend the role of two territorial factors, knowledge spillovers and related and unrelated variety, on the degree of novelty of innovation in an emerging economy.
The remainder of the paper is structured as follows. In section 2 we review the literature regarding the geography of innovation and the degree of novelty of innovation. Section 3 presents the data and the methodology employed to conduct the empirical research. Section 4 presents and discusses the results. Finally, the final section contains the conclusions and policy implications of the study.

BACKGROUND THEORY
The innovation process is strongly related to the firm's ability to absorb and transform resources linked to knowledge. It has become widely accepted that innovation is territorially embedded, in which the clustering of capabilities and resources plays an important role in stimulating innovation. Empirical evidence shows that innovative dynamics take place in relatively self-contained areas. Previous studies, based on Griliches's (1979) seminal contribution, have acknowledged the critical role of local knowledge spillovers for innovation at the regional level (Acs et al., 2002;Audretsch & Feldman, 1996;Griliches, 1979;Jaffe, 1989). In fact, proximity among players is a key factor for the assimilation and diffusion of tacit and complex knowledge, which would otherwise be unlikely and very expensive to encode (Breschi & Lissoni, 2001;Gertler, 2007;Maskell & Malmberg, 1999;Polanyi, 1966;Quatraro & Usai, 2017;Storper & Venables, 2004). Table 1 summarizes the main findings from prior studies that has examined the relationship between knowledge spillovers, related or unrelated variety and innovation.
The availability and access to knowledge sources are accordingly of paramount importance because innovation is characterized by high levels of division of labour, and firms can hardly innovate by relying only on their own internal resources (Antonelli, 2002). Knowledge flows are more easily accessible in areas characterized by highintensity innovation efforts and where actors can interact more frequently and can contact directly. The clustering of actors tends to increase opportunities for interactive learning and recombination and lower the costs of knowledge exchange, making firms more productive and innovative. Therefore, innovation does not occur in the same way everywhere, and it strongly depends on the conditions of the environment surrounding the firm. For this reason, firms in regions featuring high levels of innovation efforts are expected to enjoy advantages compared with firms located in regions with poor investments in innovation activities (Audretsch & Belitski, 2020;Carlino et al., 2006;Rodríguez-Pose & Villarreal Peralta, 2015;Schoenmakers & Duysters, 2010). In this way, assuming that innovation is strongly influenced by the characteristics of the local context, we draw the following hypothesis: Hypothesis 1: Local knowledge spillovers are positively associated with innovation.
However, it must be emphasized that not only the volume of knowledge produced matters for innovation but also its composition. Several studies based on the Schumpeterian concept of recombinant innovation have indeed investigated the determinants and effects of the intrinsic heterogeneity of regional knowledge bases (Audretsch & Belitski, 2020;Barbosa et al., 2014;Castaldi et al., 2015;Content & Frenken, 2016;Grillitsch et al., 2017;Quatraro, 2010;Schoenmakers & Duysters, 2010). R&D expenditures are comparatively more complex innovative activities than other activities related to the external acquisition of embodied technology, such as the purchase of machinery and software and the acquisition of external knowledge through technological consultancy and personnel training. Therefore, R&D expenditures can be crucial for the creation of new knowledge for innovation, and they can be the primary way of enhancing a firm's ability to assimilate and exploit existing information and generate absorptive capacity. (Antonelli & Colombelli, 2017;Audretsch & Belitski, 2020;Cohen & Levinthal, 1990b;Griffith et al., 2003;Mascarini et al., 2022).
When firms from any region increase their R&D expenditures, the region itself increases its technological capabilities and its ability to assimilate and understand new knowledge (Audretsch & Belitski, 2020;Cohen & Levinthal, 1990a;Crescenzi & Jaax, 2016). In the case of peripheral regions, or even regions in developing countries, the increase in private R&D expenditures raises the speed of incorporation of external knowledge, improving the absorptive capacity of the region and reducing the distance of local firms from the technological knowledge frontier (Gonçalves et al., 2019;Mascarini et al., 2019;Plechero & Chaminade, 2016). This is reinforced by the . Unrelated industrial variety is not significant Antonelli and Colombelli (2017) 2005-2009 Italy Regional innovation KS . The positive and significant role of external knowledge on new technological knowledge (patent) Antonietti and Gambarotto (2020) 2012-2015 Italy Regional Innovative start-ups RV UV . New start-ups are more likely where local levels of related and unrelated variety are higher . Innovative start-ups focus on the early development of breakthrough innovations and emerge where unrelated variety is higher Audretsch and Belitski (2020) 2002-2014 UK Innovation KS . Knowledge spillovers are also positively associated with the co-creation of innovation, but not for imitation or internal innovation Barbosa et al. (2014) 2004-2006 Portugal Degree of novelty of innovation KS . Extramural R&D investments appear to be the driving force of innovation for Portuguese firms . Intramural R&D intensity is not significant Basile et al. (2017) 2004-2010 Italy Start-up firms' survival RV UV . Positive relationship between related variety and firm survival in manufacturing . Positive relationship between unrelated variety and firm survival in service sectors Castaldi et al. (2015) 1977-1999 USA Regional innovation RV UV . Related variety raises the likelihood of patents . Unrelated variety raises the likelihood of breakthrough innovations Colombelli (2016) 2009-2013 Italy Regional Innovative start-ups

RV UV
. Related variety is positively associated with the creation of innovative new firms . Unrelated variety is positively correlated with the creation of innovative new firms Crescenzi and Jaax (2016) 1997-2011 Russia Regional innovation KS . The interregional knowledge flows is positive associated with regional patenting Crescenzi et al. (2007Crescenzi et al. ( ) 1990Crescenzi et al. ( -2022Crescenzi et al. ( 1990Crescenzi et al. ( -1999 Europe USA Regional innovation KS . The knowledge spillovers do not exert any statistically significant influence upon patent growth rate in the United States . The knowledge spillovers exert a positive and significant influence on territorial dynamic in the European Union Crescenzi et al. (2012Crescenzi et al. ( ) 1995Crescenzi et al. ( -2007Crescenzi et al. ( 1995Crescenzi et al. ( -2004 China India Regional innovation KS . Regional R&D spillovers is not significantly connected to regional patent in China . Regional R&D spending is important for regional innovation in India Ejdemo and Örtqvist (2020) 2008-2016 Sweden Regional innovation RV . The related variety fosters patent intensity Gonçalves et al. (2019) 2000-2011 Brazil Regional technological specialization KS . The knowledge spillover is positively associated with local technological specialization (patent) Mascarini et al. (2019) 2000-2005 Brazil Regional innovation KS . Knowledge spillovers are important for fostering regional innovation (patent) Plechero and Chaminade (2016) 2008 Degree of novelty of innovation KS . The international linkages are associated with higher degrees of novelty (product) . New-to-the-world innovation in emerging country firms is fundamentally externally driven Rodríguez-Pose and Villarreal Peralta (2015) 2000-2010 Mexico Regional growth KS . Different types of spillovers are associated the economic performance . Traditional knowledge spillovers (investments in R&D) positively affect economic growth Tavassoli and Carbonara (2014) 2002-2007 Regional innovation RV UV KS . Related variety is positively associate with patent . Unrelated variety is positive on regional innovation . The intensity of external knowledge is positive and significant Note: RV, related variety; UV, unrelated variety; KS, knowledge spillovers.
Local knowledge spillovers and the effects of related and unrelated variety on the novelty of innovation fact that in developing countries and in their regions, the main forms of innovation are through more codified ways of incorporating knowledge, especially by acquiring external knowledge embodied in machinery and software or through technological consultancy and personnel training (Arocena & Sutz, 2001;Suzigan et al., 2020;Szczygielski et al., 2017). The innovation efforts of domestic private firms generally do not include more remarkable R&D expenditures. Drivers of innovation novelty can vary for upper-and lower-level innovation. To capture these differences, which characterize the innovative dynamics in developing countries, several studies have used the notion of the degree of novelty of firms' innovation as a proxy for the outputs of innovative efforts (Amara et al., 2008;Barbosa et al., 2014;Harirchi & Chaminade, 2014;Plechero & Chaminade, 2016;Vega-Jurado et al., 2008). The focus on innovation novelty allows for assessing the extent to which the effects of different innovative efforts allow firms to achieve higher degrees of innovation outputs. To obtain upper-level innovation, firms should increase their investments in projects with higher risks since they require a longer period of development and more multifaceted and complex knowledge. Accordingly, R&D expenditure is related to upper-level innovation, which means a higher degree of novelty of innovation (Antonelli & Crespi, 2013;Berrutti & Bianchi, 2019;Szczygielski et al., 2017).
The rise of local innovative efforts in turn is a key condition for knowledge spillovers to emerge and spread across local actors. Therefore, firms localized in regions with higher R&D expenditures are exposed to higher, more complex, and diversified local knowledge spillovers that leverage the potential of innovation with higher degree of novelty. Combining this argument and the view that radical innovations are more dependent on existing knowledge than incremental innovations (Audretsch & Belitski, 2020;Barbosa et al., 2014;Schoenmakers & Duysters, 2010), we expect that upperlevel innovation would be associated with local knowledge spillovers generated from R&D expenditures. This leads us to the following hypothesis: Hypothesis 2: Local knowledge spillovers from R&D expenditures are positively associated with higher degrees of novelty of innovation.
Local knowledge spillovers can also contribute to the evolution and diversification of regions. Scholars understand that the extent of knowledge spillovers depends on the complementarity among local activities. Thus, knowledge will be shared to a larger extent as long as the cognitive distance among them is not too large (Garcia et al., 2018;Quatraro & Usai, 2017). The region's innovative performance can be affected differently by the types of variety; therefore, the variety has to be distinguished (Aarstad et al., 2016;Antonietti & Gambarotto, 2020;Castaldi et al., 2015;Frenken et al., 2007;Tavassoli & Carbonara, 2014). Related variety in a region favours the emergence of knowledge spillovers, which foster innovation. On the other hand, diversification in unrelated industries is less likely to promote such spillover impacts due to the larger technological and cognitive distance between actors in these industries. Nevertheless, this relationship should differ according to the type of variety and innovation (Castaldi et al., 2015).
Previous studies point out that spillovers between unrelated industries are less frequent than those among related industries but may be particularly conducive to radical innovation, such as the development of completely new products. In contrast, more common spillovers are between related industries but may primarily lead to incremental innovation (Boschma & Capone, 2015;Castaldi et al., 2015). Therefore, incremental innovation builds on established cognitive structures across 'related' technologies, while technological 'breakthroughs' stem from combinations across unrelated knowledge domains (Castaldi et al., 2015). This line of argument leads to the following hypotheses: Hypothesis 3: Related variety is positively associated with a lower degree of novelty of innovation.
Hypothesis 4: Unrelated variety is positively associated with a higher degree of novelty of innovation.

Data
In this section we provide information regarding the data used for the empirical analysis, followed by a discussion of the dependent variable, the variables of interest and the controls. To examine how territorial factors affect the degree of novelty of innovation in Brazil, we use a unique set of data from the Brazilian Innovation Survey (PINTEC). This dataset is maintained by the Brazilian Institute of Geography and Statistics (IBGE) and offers information on firms' innovation activities in Brazil (e.g., occurrence of innovation, expenditure, financing). It follows the Organisation for Economic Co-operation and Development's (OECD) recommendations published in the Oslo Manual. This dataset is also used by Mascarini et al. (2022). We used regional-level data from the Ministry of Labour (RAIS) to construct some territorial variables and to allocate a region (municipality or microregion) to a firm (using a code for the firm headquarters identifier). Using these two databases, we build a rather rare dataset with data on innovation at both regional and firm levels.
For this research, we used PINTEC 2011, which comprises data on Brazilian manufacturing firms for the period 2009-11. To obtain data on past innovation activities and to construct lagged explanatory variables, we had to use data from PINTEC 2008. The reason for the use of lagged variables is that innovative activities (inputs) do not immediately turn into innovation output. Hence, the dependent variable (degree of novelty of innovation) was constructed from PINTEC 2011, whereas the explanatory variables 1 are based on PINTEC and RAIS 2008.
Each sampled firm is given a code number that allows us to identify and follow each firm along the various

Dependent variable
The aim of this paper is to understand how territorial factors affect innovation novelty outcomes. In line with previous studies, our dependent variable is the degree of novelty of innovation, which is an ordinal variable (Barbosa et al., 2014;Mascarini et al., 2022;Plechero & Chaminade, 2016;Protogerou et al., 2017). The source of these data is PINTEC 2011, which asks firms to indicate whether, between 2009 and 2011, they created any product or process innovation. If the answer is positive, the firm is also asked to indicate the degree of novelty of that innovation, in which it points out among three options: new to the firm but although not new to the firm's competitors in Brazilnew-to-the-firm (1); new to domestic market but they are already available in other marketsnew-to-the-industry or domestic-market (2); and new to the global market which represents world-class innovationnew-to-the-world (3). The answer given by firms to the innovation survey sets up to our dependent variable equal to 0 when the firm did not generate innovation; equal to 1 when the firm introduced an innovation new-to-the-firm; equal to 2 when the firm introduced an innovation new-to-the-industry; and equal to 3 if the firm introduced an innovation new-to-theworld. In this way, our dependent variable distinguishes the degree of novelty of innovation at the firm level. 2 It is important to mention that most of the firms in Brazil between 2009 and 2011 did not introduce any innovations (44.8%), whereas the share of firms that introduced an innovation new-to-the-firm is 34.1%, and new-to-theindustry is 17.5%; finally, a share of 3.5% introduced world-class, new-to-the-world innovation (see Table A1 in Appendix A in the supplemental data online). Low levels of R&D expenditures translate into weak innovative performance. In addition, as in other countries, the regional distribution of innovation is uneven in Brazil (Garcia et al., 2015;Mascarini et al., 2022), and it is important to seek to understand how and why different territorial factors can make firms more prone to innovate.

Explanatory variables: Territorial factors
We analysed two territorial factors as focal explanatory variables. The first is knowledge spillovers. The literature indeed suggests that innovation activities not only improve the capacity to generate new knowledge but can also give firms the capability to internalize knowledge from other sources (Audretsch & Belitski, 2020;Barbosa et al., 2014;Breschi & Lissoni, 2001;Schoenmakers & Duysters, 2010). In this sense, knowledge spillovers can represent an important input for innovation at the firm level.
Local knowledge spillovers occur from different sources, and they vary with different innovative activities performed by local agents. R&D expenditures can generate high-intensity local knowledge spillovers. On the other hand, the incorporation of embodied technology, such as the purchase of machinery and software and the acquisition of external knowledge through technological consultancy and personnel training, has a lower capacity to generate stronger local knowledge spillovers. The difference between the local knowledge spillovers created by R&D expenditures and those produced by the acquisition of external knowledge allows us to empirically test Hypotheses 1 and 2, respectively. Therefore, we designed two different measures for local knowledge spillovers.
The first measure, used to test Hypothesis 1, involves spillovers from the total innovation efforts of local agents, including not only R&D expenses but also expenses with the incorporation of embodied technology and the acquisition of external knowledge. To this measure, we take total innovation activities by expenditures on all innovation activities divided by gross output (GO) from other firms in the same municipality and sector, SpillT m,08 . The second measure, to test Hypothesis 2, includes only knowledge spillovers from R&D efforts by taking R&D expenditures divided by gross output from other firms in municipality and sector, SpillRD m,08 .
The second territorial factor is related and unrelated variety. Following Frenken et al. (2007), we employed the entropy measure to calculate related and unrelated variety at different levels of sectoral aggregation. We use of the number of manufacturing employees in the Brazilian microregions in the five-digit Standard Industrial Classification (SIC) industries. The related and unrelated variety variables (RV m,08 and UR m,08 ) are calculated as follows: The related variety variable measures the average degree of variety of employment in five-digit SIC industries (class) with the same two-digit SIC (macro-domains). The value of the related variety ranges from 0 (employment in each macro-domain is concentrated in only one of its classes) to 3.28. 3 The unrelated variety variable measures the average degree of variety of the macro-domains. The value of unrelated variety can range from 0 (all employment is concentrated in only one macro-domain) to 5 (all macrodomains employ an equal number of employees) (Boschma & Iammarino, 2007;Frenken et al., 2007;Krafft et al., 2014). 4 We include regional-level control variables, such as local human capital, relative wealth, and a dummy for macro-regions. We considered two different measures to capture the human capital in a firm's municipality: the qualification of employees measured by the percentage of employees with higher education (graduated, master's and doctorate) in the firm's municipality -Educ m,08 and the resources in science and technology (S&T) measured by the percentage of the workforce employed in technological and science occupations in the firm's municipality, ST m,08 . Relative wealth reflects a region's overall level of development and is proxied by gross domestic product (GDP) per capita for the firm's municipality, GDP m,08 . We also used a dummy with six macro-regions: São Paulo state (SP), North, Northeast, Southeast without SP, South and Midwest. We separated São Paulo state from the macro region of the Southeast due to the strong regional concentration of innovation in that Brazilian State. As in previous studies we include control variables that are usually considered to influence innovation at the firm level, such as innovation expenditure, size, productivity, dummies to sectoral industries, public funding, collaborations and the ownership (Crescenzi & Rodríguez-Pose, 2017;Mascarini et al., 2019Mascarini et al., , 2022. We even include the foreign ownership due to the growing recognition of the role of multinational corporations (MNCs) in regional innovation, especially in developing countries. Previous studies show that the entry of MNCs into a region has positive effects on regional innovation, as it can represent an important source of novelty for local producers (Crescenzi & Iammarino, 2017;Ascani et al., 2020). Finally, we used Castellacci taxonomy and added the extractive industry for dummies to sectorial industries (Castellacci, 2008;De Fuentes et al., 2015), this aggregation is more appropriated to analysis, because an excessive disaggregation of the dummies would cause econometric problems given the implemented estimator, Table 2 summarizes all the variables. For the correlation matrix, see Table A2 in Appendix A in the supplemental data online.

The effect of knowledge spillovers on the degree of novelty of innovation
To explore the relationship between the degree of novelty of the innovation and territorial factors, we apply the generalized form of the ordered probit model (OPM). This approach allows the estimation of different coefficients for distinct categories and is specified in cases in which the explanatory variable is categorical and the proportional odds assumption is violated. The benefit of using a generalized ordered probit model (GOPM) over a standard OPM is that the information covered in the ordinal explanatory variable can be used without the restriction of parallel regressions for the distinct categories of the dependent variable. The parallel regression assumption, sometimes described as the proportional odds assumption, of an OPM is where all βj values are restricted to be equal across the different categories of the dependent variable. Relaxing this restriction is warranted because the explanatory variables may not affect all groups equally (Williams, 2016). Each model gives us three columns of estimates of the binary models, since we have four categories in our dependent variable (did not innovate, 0; new-to-the-firm, 1; new-to-the-industry, 2; and new-to-the-world, 3). The first column contrasts firms that did not innovate (dependent variable equalling 0) with firms that did innovate (dependent variable > 0) between 2009 and 2011. The second column contrasts firms that did not innovate or the innovation is new-to-the-firm (dependent variable equal to 0 or 1) with firms that introduced innovation new-to-the-industry or new-to-the-world (dependent variable equal to 2 or 3). The last column contrasts firms that introduced innovation new-to-the-world (dependent variable equal to 3) with other firms (dependent variable < 3).
Because we use two different measures for regional knowledge spillovers and qualified human capital, we built four sets of estimations. Models using the local human capital measured by the percentage of employment with higher education (Educ) are presented as models I and II, and those using the share of employment in science and technology (ST) are models III and IV. Similarly, the knowledge spillovers generated from total innovation expenditures are added in models I and III, and those from R&D expenditures are included in models II and IV (Table 3).
Our results show that territorial factors are especially important if a firm wants to move from new-to-the-firm innovation to a higher degree of novelty of innovation (column 2 in the models, Table 3). There is a positive association between knowledge spillovers from total innovative activities (SpillT) and innovations (models I and III, Table 3), confirming both theoretical expectations and previous empirical studies. Additionally, the coefficient is not significant (columns 2 and 3) in models I and III. These results support Hypothesis 1 and suggest that knowledge spillovers are especially important to generate innovation but are not as valuable as the degree of novelty innovation increases. Therefore, an increase in the innovation expenditures of neighbouring firms (in the same municipality) may increase a firm's likelihood of innovation.
In contrast, the coefficient of regional knowledge spillovers from R&D expenditures (SpillRD) is positive and significant only in column 3 in models II and IV. This result shows that R&D efforts are important for new-tothe-world innovations, providing support for Hypothesis 2. This finding suggests that as neighbouring firms' R&D expenditures increase, it is more likely that the firm's innovation will be of higher novelty. Therefore, firms in regions with stronger knowledge spillovers from R&D expenditures are more likely to launch novel and upper-level innovations than firms in other regions. Thus, locating in areas where there are more firms with R&D expenditures may be a strategy for firms that want to introduce innovations to the world. This finding represents an important contribution of our paper to the literature, since there was a gap in studies that perceived the main effects of the different types of spillovers on the degree of novelty of innovation.
These results help us to understand that distinct regional knowledge spillovers generated by the innovation expenditures of private firms can foster different degrees of novelty of innovation. In this way, our findings indicate that regional knowledge spillovers based on total innovation activities, not only R&D expenditures, are supportive of a critical mass of innovation, while regional knowledge spillovers based on R&D provide upper-level innovation with a higher degree of novelty.

Variety and innovation novelty
Regarding variety, both related and unrelated variety are associated with the degree of novelty of innovation. The coefficient of related variety is negative and significant, while the coefficient of unrelated variety is positive and significant in column 2 (all models, Table 3). The negative sign for related variety is unexpected and suggests that firms in regions with higher regional-related variety are less likely to move on of new-to-the-market or new-tothe-world innovation. Therefore, this result does not support Hypothesis 3 and suggests that economies and capabilities arising from related variety do not seem to be associated with upper-level innovation. In some cases, this context may hamper moving from no innovation and new-to-the-firm innovation (0 and 1 values) to a higher degree of novelty (2 and 3 values). Our result is opposite to those of previous studies, which showed that related variety in a region is favourable to the emergence of knowledge spillovers that generate regional innovation (Table 1).
A possible explanation for this result is that in developing countries, such as Brazil, whose regions are characterized by low diversification, an increase in related variety could mean the process in which actors in a region engage in industries with very similar knowledge bases that could downfall Jacob's externalities. Our measure of related  variety shows that each sector draws on specific sets of capabilities and technical knowledge. Therefore, it is reasonable to expect that too much specialization (related variety) does not help to get out of the established and familiar technological trajectories. After all, the distinction between the three kinds of innovation maps on the degree to which firms may act as pioneers in the adoption of a product or process innovation. Therefore, we find that adopters (or imitators) are influenced by the local structure of economic activities only in terms of intermediate levels of originality.
In contrast, the positive sign for unrelated variety supports the hypothesis that regions with more diversified activities are also regions where firms have a higher degree of innovation, supporting Hypothesis 4. Thus, localization in regions with higher concentrations of related subsectors and a more even spread across unrelated sectors may be associated with more novel innovation. This result is convergent with the previous literature (Antonietti & Gambarotto, 2020;Castaldi et al., 2015;Hesse & Fornahl, 2020) and shows the importance of diversity in the sources of novelty for innovation, knowledge and learning, which can come from a diverse set of economic activities and technological capabilities. Urban economies therefore support a greater degree of novelty, particularly for firms moving from no innovation and new-to-the-firm innovation (0 and 1 values) to new-to-the-industry and new-to-the-world innovation (2 and 3 values). This finding, combined with the results of related variety, reinforces the idea of the importance of diversification to foster innovation, especially for innovations new-to-the-industry.
Regarding the controls, the coefficients for the qualification of the workforce are positive and significant across the highest degrees of novelty (columns 2 and 3, model I). This finding suggests that a firm's innovation output is likely to have greater novelty when firms are in regions with more local qualified human capital. Therefore, regions with a higher availability of high-skilled employees become significant in relation to innovation as the degree of innovation increases. In addition, the coefficient of S&T resources is positive and significant in column 2 in models III and IV, which provides evidence that greater availability of S&T resources in regions increases the likelihood of moving from no innovation and new-to-the-firm innovation (0 and 1 values) to new-to-the-industry and new-to-the-world innovation (2 and 3 values). The results show not only that internal human capital leads to greater degrees of novelty but also that external human capital does. Therefore, human capital is a key factor for innovation, and the clustering of technological capabilities and skilled workers fosters the creation and diffusion of knowledge for innovation with higher degrees of novelty.
We also identify a positive and statistically significant association between GDP per capita and the degree of novelty in columns 2 and 3 in all models in Table 3. This finding indicates that firms located in more developed regions are more likely to generate more upperlevel innovations, especially new-to-the-world innovations, than firms located elsewhere. This provides an indication Local knowledge spillovers and the effects of related and unrelated variety on the novelty of innovation 1675 Table 4. Marginal effects based on generalized ordered probit model (GOPM)degree of novelty of innovation. that how well off a region is in comparison to the national average can be a differential source of competitive advantage for firms to promote more novel innovation. Therefore, it is more likely that the firm's innovation output will be of higher novelty in regions with a higher GDP per capita. Finally, the coefficient for the regional dummy for the North is negative and significant, indicating a lower degree of novelty introduced by firms in the North than in São Paulo state. Similarly, the coefficient for Midwest in column 3 (all models, Table 3) is negative and significant, meaning that less new-to-the-world innovation was generated by firms in the Midwest, again compared with São Paulo state. On the other hand, the coefficient for the South dummy in the first column is positive and significant, which suggests more new-to-the-firm innovation in the southern region of the country than in São Paulo state.

Variable
Similar to other nonlinear models, we can estimate the marginal effects of explanatory variablesterritorial factorson the probabilities. By doing so, we aim to capture the magnitude of the effects within each degree of novelty.
The results show that regarding knowledge spillovers (SpillT), the coefficient is negative for no innovation and positive for new-to-the-firm innovation (columns 0 and 1 in models I and III, Table 4). A 1% increase in the logarithm of innovation expenditures of neighbouring firms (in the same municipality) may increase a firm's likelihood of undertaking innovation itself (new-to-the-firm). Since the percentage of innovations new-to-the-firms is 34.1% (2450/7181), the effect of 0.8 percentage points represents the likelihood of rising approximately 59 innovations newto-the-firm.
In addition, the coefficient of marginal effects for spillovers from R&D expenditures (SpillRD) is positive and significant only in the last column (3) in models II and IV, which suggests that R&D spillovers particularly impact radical innovation. In terms of magnitude, a 1% increase in the logarithm of R&D efforts of neighbouring firms raises the probability of a firm introducing new-tothe-world innovation by approximately 0.17-0.23 percentage points. Note that 0.17 percentage points represents the likelihood rise approximately 12 innovations new-tothe-word, since this is 3.57% (256/7181).
Evaluating the marginal effect on related and unrelated variety, we find that the coefficients are significant only in column 2 (Table 4), and a 1% increase in related variety decreases the probability of a firm introducing new-to-theindustry innovations by approximately 0.07 percentage points. A 1% increase in unrelated variety raises the probability that a firm introduces new-to-the-industry innovation by approximately 0.04 percentage points, representing three innovations new-to-the-industry. The likelihood of a firm introducing new-to-the-industry or new-to-the-world innovations is increased due to increases in GDP per capita by 0.05 and 0.01 percentage points, respectively.
Regarding local qualified human capital, a 1% increase in the regional share of employees with higher education (Educ) raises the probability of a firm generating new-to-the-world innovations by approximately 0.07 percentage points. A 1% increase in the regional share of employment in S&T (ST) decreases the probability of a firm creating new-to-the-firm innovations by approximately 1.2 percentage points and raises the probability of a firm introducing new-to-the-industry innovations by approximately 2.0 percentage points.

CONCLUSIONS AND POLICY IMPLICATIONS
The relationship between innovation and territory has been the subject of increasing attention in the literature. Several studies show that territorial factors affect innovation in regions, emphasizing the role of local knowledge spillovers. However, this literature has paid little attention to how territorial factors affect innovation at the firm level, especially their impacts on the degree of novelty of innovation in developing countries. Thus, this study sought to address this research gap by deepening current knowledge about the role of local knowledge spillovers and the effects of related and unrelated variety as determinants of different degrees of novelty of innovation. In addition, this paper aligns with a growing body of literature that aims to contribute to increasing the understanding of the analysis of territorial dynamics in emerging countries. We use a unique dataset on innovation at the regional and firm levels, and we implement our empirical strategy in an emerging country, such as Brazil.
We used GOPM to estimate the coefficients for distinct degrees of novelty of innovation. Our results show that even in developing countries, where the degree of innovation is low and characterized mostly by imitation strategies, territorial factors play an important role in stimulating firms to upgrade by undertaking more novel forms of innovation. Local knowledge spillovers generated from R&D expenditures positively affect the likelihood of local firms introducing new-to-the-world innovations. Instead, spillovers generated from total innovation expenditures stimulate local firms to introduce innovation, but especially the new-to-the-firm. These results not only show the importance of territorial factors for the degree of novelty of innovation but also highlight how spillovers foster different levels of innovation. Regarding the related and unrelated variety, our findings show that both are associated with new-to-theindustry innovations, but in the opposite way. Whereas increasing related variety decreases the probability of firm generating new-to-the-industry innovation, raising unrelated variety increases the probability of firm introduces new-to-the-industry innovation. In other words, a firm is more likely to generate new-to-the-industry innovations the greater the unrelated variety or smaller the related variety of the region where the firm is located. This result shows the importance of diverse sources of innovation, especially for upper-level innovations. This finding is reinforced by the relationship between newto-the-world innovations and the presence of higher education workers, neighbouring firms with R&D effort and more developed regions.
This research represents an initial step towards investigating the direct linkages between territorial factors and the degree of novelty of innovation in Brazil. Nevertheless, this study has some limitations, such as its cross-sectional nature, which gives us only a snapshot of the influence of territorial factors on the degree of novelty of innovation. Future research based on panel data could complement this analysis by applying structural and dynamic modelling. However, our results can shed new light on the debate regarding the role of territorial factors in developing countries, which tend to have a low degree of innovation and strong heterogeneity among regions.
These findings have some policy implications. To reinforce the positive effects of territorial factors, it is necessary to design different policy strategies for regions with different characteristics. More developed regions require incentives to undertake new-to-the-world innovations; policies should design mechanisms and instruments that increase local firms' R&D expenditures because of their positive impacts on generating local knowledge spillovers and new-to-the-world innovations. In cases where the main forms of innovation are incremental, policies can stimulate general innovation expenditures since they have a larger impact on generating local knowledge spillovers and new-to-the-firm innovations. Another way to encourage local firms to introduce new-to-the-world innovations, especially for local policymakers, is to attract highly qualified workers since they can generate local knowledge spillovers that can foster new-to-the-world innovations. In addition, policies that attract S&T workers can increase the chance that firms generate new-to-the-industry innovation. Providing new sources of novelty for innovation should also be an important aim for local policymakers since new-to-the-industry innovation is positively associated with less related variety and more unrelated variety.

DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors. Degree = y if PRODNOV = y AND PROCNOV ≤ y OR if PRODNOV ≤ y AND PROCNOV = y; y = 0, 1, 2, 3.

FUNDING
3. Since our empirical analysis is based on 310 classes (five-digit industries -D) within 32 macro-domains (two-digit industries -G), the maximum limit for related variety is log 2 (D) − log 2 (G). 4. Since the maximum limit for unrelated variety is log 2 G and G ¼ 32.